fix: list multiple @mentions

fix: simplify @llm warning
feat: 展示@agent引用
2026-01-16 00:19:28 +08:00 · 2026-01-16 00:08:51 +08:00 · 2026-01-15 23:32:14 +08:00 · 2026-01-15 17:53:28 +08:00 · 2026-01-15 17:12:30 +08:00 · 2026-01-15 16:44:15 +08:00
594 changed files with 49110 additions and 19417 deletions
@@ -0,0 +1 @@
+../.claude/skills
@@ -5,5 +5,18 @@
    "typescript-lsp@claude-plugins-official": true,
    "pyright-lsp@claude-plugins-official": true,
    "ralph-loop@claude-plugins-official": true
+  },
+  "hooks": {
+    "PreToolUse": [
+      {
+        "matcher": "Bash",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "npx -y [email protected]"
+          }
+        ]
+      }
+    ]
  }
 }
@@ -0,0 +1,46 @@
+---
+name: orpc-contract-first
+description: Guide for implementing oRPC contract-first API patterns in Dify frontend. Triggers when creating new API contracts, adding service endpoints, integrating TanStack Query with typed contracts, or migrating legacy service calls to oRPC. Use for all API layer work in web/contract and web/service directories.
+---
+
+# oRPC Contract-First Development
+
+## Project Structure
+
+```
+web/contract/
+├── base.ts           # Base contract (inputStructure: 'detailed')
+├── router.ts         # Router composition & type exports
+├── marketplace.ts    # Marketplace contracts
+└── console/          # Console contracts by domain
+    ├── system.ts
+    └── billing.ts
+```
+
+## Workflow
+
+1. **Create contract** in `web/contract/console/{domain}.ts`
+   - Import `base` from `../base` and `type` from `@orpc/contract`
+   - Define route with `path`, `method`, `input`, `output`
+
+2. **Register in router** at `web/contract/router.ts`
+   - Import directly from domain file (no barrel files)
+   - Nest by API prefix: `billing: { invoices, bindPartnerStack }`
+
+3. **Create hooks** in `web/service/use-{domain}.ts`
+   - Use `consoleQuery.{group}.{contract}.queryKey()` for query keys
+   - Use `consoleClient.{group}.{contract}()` for API calls
+
+## Key Rules
+
+- **Input structure**: Always use `{ params, query?, body? }` format
+- **Path params**: Use `{paramName}` in path, match in `params` object
+- **Router nesting**: Group by API prefix (e.g., `/billing/*` → `billing: {}`)
+- **No barrel files**: Import directly from specific files
+- **Types**: Import from `@/types/`, use `type<T>()` helper
+
+## Type Export
+
+```typescript
+export type ConsoleInputs = InferContractRouterInputs<typeof consoleRouterContract>
+```
@@ -39,12 +39,6 @@ jobs:
      - name: Install dependencies
        run: uv sync --project api --dev

-      - name: Run pyrefly check
-        run: |
-          cd api
-          uv add --dev pyrefly
-          uv run pyrefly check || true
-
      - name: Run dify config tests
        run: uv run --project api dev/pytest/pytest_config_tests.py

@@ -1,4 +1,4 @@
-name: Deploy Trigger Dev
+name: Deploy Agent Dev

 permissions:
  contents: read
@@ -7,7 +7,7 @@ on:
  workflow_run:
    workflows: ["Build and Push API & Web"]
    branches:
-      - "deploy/trigger-dev"
+      - "deploy/agent-dev"
    types:
      - completed

@@ -16,12 +16,12 @@ jobs:
    runs-on: ubuntu-latest
    if: |
      github.event.workflow_run.conclusion == 'success' &&
-      github.event.workflow_run.head_branch == 'deploy/trigger-dev'
+      github.event.workflow_run.head_branch == 'deploy/agent-dev'
    steps:
      - name: Deploy to server
        uses: appleboy/[email protected]
        with:
-          host: ${{ secrets.TRIGGER_SSH_HOST }}
+          host: ${{ secrets.AGENT_DEV_SSH_HOST }}
          username: ${{ secrets.SSH_USER }}
          key: ${{ secrets.SSH_PRIVATE_KEY }}
          script: |
@@ -0,0 +1,29 @@
+name: Deploy HITL
+
+on:
+  workflow_run:
+    workflows: ["Build and Push API & Web"]
+    branches:
+      - "feat/hitl-frontend"
+      - "feat/hitl-backend"
+    types:
+      - completed
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    if: |
+      github.event.workflow_run.conclusion == 'success' &&
+      (
+        github.event.workflow_run.head_branch == 'feat/hitl-frontend' ||
+        github.event.workflow_run.head_branch == 'feat/hitl-backend'
+      )
+    steps:
+      - name: Deploy to server
+        uses: appleboy/[email protected]
+        with:
+          host: ${{ secrets.HITL_SSH_HOST }}
+          username: ${{ secrets.SSH_USER }}
+          key: ${{ secrets.SSH_PRIVATE_KEY }}
+          script: |
+            ${{ vars.SSH_SCRIPT || secrets.SSH_SCRIPT }}
@@ -90,7 +90,7 @@ jobs:
        uses: actions/setup-node@v6
        if: steps.changed-files.outputs.any_changed == 'true'
        with:
-          node-version: 22
+          node-version: 24
          cache: pnpm
          cache-dependency-path: ./web/pnpm-lock.yaml

@@ -16,10 +16,6 @@ jobs:
    name: unit test for Node.js SDK
    runs-on: ubuntu-latest

-    strategy:
-      matrix:
-        node-version: [16, 18, 20, 22]
-
    defaults:
      run:
        working-directory: sdks/nodejs-client
@@ -29,10 +25,10 @@ jobs:
        with:
          persist-credentials: false

-      - name: Use Node.js ${{ matrix.node-version }}
+      - name: Use Node.js
        uses: actions/setup-node@v6
        with:
-          node-version: ${{ matrix.node-version }}
+          node-version: 24
          cache: ''
          cache-dependency-path: 'pnpm-lock.yaml'

@@ -1,94 +0,0 @@
-name: Translate i18n Files Based on English
-
-on:
-  push:
-    branches: [main]
-    paths:
-      - 'web/i18n/en-US/*.json'
-  workflow_dispatch:
-
-permissions:
-  contents: write
-  pull-requests: write
-
-jobs:
-  check-and-update:
-    if: github.repository == 'langgenius/dify'
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        working-directory: web
-    steps:
-      # Keep use old checkout action version for https://github.com/peter-evans/create-pull-request/issues/4272
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          token: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Check for file changes in i18n/en-US
-        id: check_files
-        run: |
-          # Skip check for manual trigger, translate all files
-          if [ "${{ github.event_name }}" == "workflow_dispatch" ]; then
-            echo "FILES_CHANGED=true" >> $GITHUB_ENV
-            echo "FILE_ARGS=" >> $GITHUB_ENV
-            echo "Manual trigger: translating all files"
-          else
-            git fetch origin "${{ github.event.before }}" || true
-            git fetch origin "${{ github.sha }}" || true
-            changed_files=$(git diff --name-only "${{ github.event.before }}" "${{ github.sha }}" -- 'i18n/en-US/*.json')
-            echo "Changed files: $changed_files"
-            if [ -n "$changed_files" ]; then
-              echo "FILES_CHANGED=true" >> $GITHUB_ENV
-              file_args=""
-              for file in $changed_files; do
-                filename=$(basename "$file" .json)
-                file_args="$file_args --file $filename"
-              done
-              echo "FILE_ARGS=$file_args" >> $GITHUB_ENV
-              echo "File arguments: $file_args"
-            else
-              echo "FILES_CHANGED=false" >> $GITHUB_ENV
-            fi
-          fi
-
-      - name: Install pnpm
-        uses: pnpm/action-setup@v4
-        with:
-          package_json_file: web/package.json
-          run_install: false
-
-      - name: Set up Node.js
-        if: env.FILES_CHANGED == 'true'
-        uses: actions/setup-node@v6
-        with:
-          node-version: 'lts/*'
-          cache: pnpm
-          cache-dependency-path: ./web/pnpm-lock.yaml
-
-      - name: Install dependencies
-        if: env.FILES_CHANGED == 'true'
-        working-directory: ./web
-        run: pnpm install --frozen-lockfile
-
-      - name: Generate i18n translations
-        if: env.FILES_CHANGED == 'true'
-        working-directory: ./web
-        run: pnpm run i18n:gen ${{ env.FILE_ARGS }}
-
-      - name: Create Pull Request
-        if: env.FILES_CHANGED == 'true'
-        uses: peter-evans/create-pull-request@v6
-        with:
-          token: ${{ secrets.GITHUB_TOKEN }}
-          commit-message: 'chore(i18n): update translations based on en-US changes'
-          title: 'chore(i18n): translate i18n files based on en-US changes'
-          body: |
-            This PR was automatically created to update i18n translation files based on changes in en-US locale.
-
-            **Triggered by:** ${{ github.sha }}
-
-            **Changes included:**
-            - Updated translation files for all locales
-          branch: chore/automated-i18n-updates-${{ github.sha }}
-          delete-branch: true
@@ -0,0 +1,421 @@
+name: Translate i18n Files with Claude Code
+
+# Note: claude-code-action doesn't support push events directly.
+# Push events are handled by trigger-i18n-sync.yml which sends repository_dispatch.
+# See: https://github.com/langgenius/dify/issues/30743
+
+on:
+  repository_dispatch:
+    types: [i18n-sync]
+  workflow_dispatch:
+    inputs:
+      files:
+        description: 'Specific files to translate (space-separated, e.g., "app common"). Leave empty for all files.'
+        required: false
+        type: string
+      languages:
+        description: 'Specific languages to translate (space-separated, e.g., "zh-Hans ja-JP"). Leave empty for all supported languages.'
+        required: false
+        type: string
+      mode:
+        description: 'Sync mode: incremental (only changes) or full (re-check all keys)'
+        required: false
+        default: 'incremental'
+        type: choice
+        options:
+          - incremental
+          - full
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  translate:
+    if: github.repository == 'langgenius/dify'
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Configure Git
+        run: |
+          git config --global user.name "github-actions[bot]"
+          git config --global user.email "github-actions[bot]@users.noreply.github.com"
+
+      - name: Install pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          package_json_file: web/package.json
+          run_install: false
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v6
+        with:
+          node-version: 24
+          cache: pnpm
+          cache-dependency-path: ./web/pnpm-lock.yaml
+
+      - name: Detect changed files and generate diff
+        id: detect_changes
+        run: |
+          if [ "${{ github.event_name }}" == "workflow_dispatch" ]; then
+            # Manual trigger
+            if [ -n "${{ github.event.inputs.files }}" ]; then
+              echo "CHANGED_FILES=${{ github.event.inputs.files }}" >> $GITHUB_OUTPUT
+            else
+              # Get all JSON files in en-US directory
+              files=$(ls web/i18n/en-US/*.json 2>/dev/null | xargs -n1 basename | sed 's/.json$//' | tr '\n' ' ')
+              echo "CHANGED_FILES=$files" >> $GITHUB_OUTPUT
+            fi
+            echo "TARGET_LANGS=${{ github.event.inputs.languages }}" >> $GITHUB_OUTPUT
+            echo "SYNC_MODE=${{ github.event.inputs.mode || 'incremental' }}" >> $GITHUB_OUTPUT
+
+            # For manual trigger with incremental mode, get diff from last commit
+            # For full mode, we'll do a complete check anyway
+            if [ "${{ github.event.inputs.mode }}" == "full" ]; then
+              echo "Full mode: will check all keys" > /tmp/i18n-diff.txt
+              echo "DIFF_AVAILABLE=false" >> $GITHUB_OUTPUT
+            else
+              git diff HEAD~1..HEAD -- 'web/i18n/en-US/*.json' > /tmp/i18n-diff.txt 2>/dev/null || echo "" > /tmp/i18n-diff.txt
+              if [ -s /tmp/i18n-diff.txt ]; then
+                echo "DIFF_AVAILABLE=true" >> $GITHUB_OUTPUT
+              else
+                echo "DIFF_AVAILABLE=false" >> $GITHUB_OUTPUT
+              fi
+            fi
+          elif [ "${{ github.event_name }}" == "repository_dispatch" ]; then
+            # Triggered by push via trigger-i18n-sync.yml workflow
+            # Validate required payload fields
+            if [ -z "${{ github.event.client_payload.changed_files }}" ]; then
+              echo "Error: repository_dispatch payload missing required 'changed_files' field" >&2
+              exit 1
+            fi
+            echo "CHANGED_FILES=${{ github.event.client_payload.changed_files }}" >> $GITHUB_OUTPUT
+            echo "TARGET_LANGS=" >> $GITHUB_OUTPUT
+            echo "SYNC_MODE=${{ github.event.client_payload.sync_mode || 'incremental' }}" >> $GITHUB_OUTPUT
+
+            # Decode the base64-encoded diff from the trigger workflow
+            if [ -n "${{ github.event.client_payload.diff_base64 }}" ]; then
+              if ! echo "${{ github.event.client_payload.diff_base64 }}" | base64 -d > /tmp/i18n-diff.txt 2>&1; then
+                echo "Warning: Failed to decode base64 diff payload" >&2
+                echo "" > /tmp/i18n-diff.txt
+                echo "DIFF_AVAILABLE=false" >> $GITHUB_OUTPUT
+              elif [ -s /tmp/i18n-diff.txt ]; then
+                echo "DIFF_AVAILABLE=true" >> $GITHUB_OUTPUT
+              else
+                echo "DIFF_AVAILABLE=false" >> $GITHUB_OUTPUT
+              fi
+            else
+              echo "" > /tmp/i18n-diff.txt
+              echo "DIFF_AVAILABLE=false" >> $GITHUB_OUTPUT
+            fi
+          else
+            echo "Unsupported event type: ${{ github.event_name }}"
+            exit 1
+          fi
+
+          # Truncate diff if too large (keep first 50KB)
+          if [ -f /tmp/i18n-diff.txt ]; then
+            head -c 50000 /tmp/i18n-diff.txt > /tmp/i18n-diff-truncated.txt
+            mv /tmp/i18n-diff-truncated.txt /tmp/i18n-diff.txt
+          fi
+
+          echo "Detected files: $(cat $GITHUB_OUTPUT | grep CHANGED_FILES || echo 'none')"
+
+      - name: Run Claude Code for Translation Sync
+        if: steps.detect_changes.outputs.CHANGED_FILES != ''
+        uses: anthropics/claude-code-action@v1
+        with:
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          prompt: |
+            You are a professional i18n synchronization engineer for the Dify project.
+            Your task is to keep all language translations in sync with the English source (en-US).
+
+            ## CRITICAL TOOL RESTRICTIONS
+            - Use **Read** tool to read files (NOT cat or bash)
+            - Use **Edit** tool to modify JSON files (NOT node, jq, or bash scripts)
+            - Use **Bash** ONLY for: git commands, gh commands, pnpm commands
+            - Run bash commands ONE BY ONE, never combine with && or ||
+            - NEVER use `$()` command substitution - it's not supported. Split into separate commands instead.
+
+            ## WORKING DIRECTORY & ABSOLUTE PATHS
+            Claude Code sandbox working directory may vary. Always use absolute paths:
+            - For pnpm: `pnpm --dir ${{ github.workspace }}/web <command>`
+            - For git: `git -C ${{ github.workspace }} <command>`
+            - For gh: `gh --repo ${{ github.repository }} <command>`
+            - For file paths: `${{ github.workspace }}/web/i18n/`
+
+            ## EFFICIENCY RULES
+            - **ONE Edit per language file** - batch all key additions into a single Edit
+            - Insert new keys at the beginning of JSON (after `{`), lint:fix will sort them
+            - Translate ALL keys for a language mentally first, then do ONE Edit
+
+            ## Context
+            - Changed/target files: ${{ steps.detect_changes.outputs.CHANGED_FILES }}
+            - Target languages (empty means all supported): ${{ steps.detect_changes.outputs.TARGET_LANGS }}
+            - Sync mode: ${{ steps.detect_changes.outputs.SYNC_MODE }}
+            - Translation files are located in: ${{ github.workspace }}/web/i18n/{locale}/{filename}.json
+            - Language configuration is in: ${{ github.workspace }}/web/i18n-config/languages.ts
+            - Git diff is available: ${{ steps.detect_changes.outputs.DIFF_AVAILABLE }}
+
+            ## CRITICAL DESIGN: Verify First, Then Sync
+
+            You MUST follow this three-phase approach:
+
+            ═══════════════════════════════════════════════════════════════
+            ║  PHASE 1: VERIFY - Analyze and Generate Change Report       ║
+            ═══════════════════════════════════════════════════════════════
+
+            ### Step 1.1: Analyze Git Diff (for incremental mode)
+            Use the Read tool to read `/tmp/i18n-diff.txt` to see the git diff.
+
+            Parse the diff to categorize changes:
+            - Lines with `+` (not `+++`): Added or modified values
+            - Lines with `-` (not `---`): Removed or old values
+            - Identify specific keys for each category:
+              * ADD: Keys that appear only in `+` lines (new keys)
+              * UPDATE: Keys that appear in both `-` and `+` lines (value changed)
+              * DELETE: Keys that appear only in `-` lines (removed keys)
+
+            ### Step 1.2: Read Language Configuration
+            Use the Read tool to read `${{ github.workspace }}/web/i18n-config/languages.ts`.
+            Extract all languages with `supported: true`.
+
+            ### Step 1.3: Run i18n:check for Each Language
+            ```bash
+            pnpm --dir ${{ github.workspace }}/web install --frozen-lockfile
+            ```
+            ```bash
+            pnpm --dir ${{ github.workspace }}/web run i18n:check
+            ```
+
+            This will report:
+            - Missing keys (need to ADD)
+            - Extra keys (need to DELETE)
+
+            ### Step 1.4: Generate Change Report
+
+            Create a structured report identifying:
+            ```
+            ╔══════════════════════════════════════════════════════════════╗
+            ║                    I18N SYNC CHANGE REPORT                   ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ Files to process: [list]                                     ║
+            ║ Languages to sync: [list]                                    ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ ADD (New Keys):                                              ║
+            ║   - [filename].[key]: "English value"                        ║
+            ║   ...                                                        ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ UPDATE (Modified Keys - MUST re-translate):                  ║
+            ║   - [filename].[key]: "Old value" → "New value"              ║
+            ║   ...                                                        ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ DELETE (Extra Keys):                                         ║
+            ║   - [language]/[filename].[key]                              ║
+            ║   ...                                                        ║
+            ╚══════════════════════════════════════════════════════════════╝
+            ```
+
+            **IMPORTANT**: For UPDATE detection, compare git diff to find keys where
+            the English value changed. These MUST be re-translated even if target
+            language already has a translation (it's now stale!).
+
+            ═══════════════════════════════════════════════════════════════
+            ║  PHASE 2: SYNC - Execute Changes Based on Report            ║
+            ═══════════════════════════════════════════════════════════════
+
+            ### Step 2.1: Process ADD Operations (BATCH per language file)
+
+            **CRITICAL WORKFLOW for efficiency:**
+            1. First, translate ALL new keys for ALL languages mentally
+            2. Then, for EACH language file, do ONE Edit operation:
+               - Read the file once
+               - Insert ALL new keys at the beginning (right after the opening `{`)
+               - Don't worry about alphabetical order - lint:fix will sort them later
+
+            Example Edit (adding 3 keys to zh-Hans/app.json):
+            ```
+            old_string: '{\n  "accessControl"'
+            new_string: '{\n  "newKey1": "translation1",\n  "newKey2": "translation2",\n  "newKey3": "translation3",\n  "accessControl"'
+            ```
+
+            **IMPORTANT**:
+            - ONE Edit per language file (not one Edit per key!)
+            - Always use the Edit tool. NEVER use bash scripts, node, or jq.
+
+            ### Step 2.2: Process UPDATE Operations
+
+            **IMPORTANT: Special handling for zh-Hans and ja-JP**
+            If zh-Hans or ja-JP files were ALSO modified in the same push:
+            - Run: `git -C ${{ github.workspace }} diff HEAD~1 --name-only` and check for zh-Hans or ja-JP files
+            - If found, it means someone manually translated them. Apply these rules:
+
+            1. **Missing keys**: Still ADD them (completeness required)
+            2. **Existing translations**: Compare with the NEW English value:
+               - If translation is **completely wrong** or **unrelated** → Update it
+               - If translation is **roughly correct** (captures the meaning) → Keep it, respect manual work
+               - When in doubt, **keep the manual translation**
+
+            Example:
+            - English changed: "Save" → "Save Changes"
+            - Manual translation: "保存更改" → Keep it (correct meaning)
+            - Manual translation: "删除" → Update it (completely wrong)
+
+            For other languages:
+            Use Edit tool to replace the old value with the new translation.
+            You can batch multiple updates in one Edit if they are adjacent.
+
+            ### Step 2.3: Process DELETE Operations
+            For extra keys reported by i18n:check:
+            - Run: `pnpm --dir ${{ github.workspace }}/web run i18n:check --auto-remove`
+            - Or manually remove from target language JSON files
+
+            ## Translation Guidelines
+
+            - PRESERVE all placeholders exactly as-is:
+              - `{{variable}}` - Mustache interpolation
+              - `${variable}` - Template literal
+              - `<tag>content</tag>` - HTML tags
+              - `_one`, `_other` - Pluralization suffixes (these are KEY suffixes, not values)
+            - Use appropriate language register (formal/informal) based on existing translations
+            - Match existing translation style in each language
+            - Technical terms: check existing conventions per language
+            - For CJK languages: no spaces between characters unless necessary
+            - For RTL languages (ar-TN, fa-IR): ensure proper text handling
+
+            ## Output Format Requirements
+            - Alphabetical key ordering (if original file uses it)
+            - 2-space indentation
+            - Trailing newline at end of file
+            - Valid JSON (use proper escaping for special characters)
+
+            ═══════════════════════════════════════════════════════════════
+            ║  PHASE 3: RE-VERIFY - Confirm All Issues Resolved           ║
+            ═══════════════════════════════════════════════════════════════
+
+            ### Step 3.1: Run Lint Fix (IMPORTANT!)
+            ```bash
+            pnpm --dir ${{ github.workspace }}/web lint:fix --quiet -- 'i18n/**/*.json'
+            ```
+            This ensures:
+            - JSON keys are sorted alphabetically (jsonc/sort-keys rule)
+            - Valid i18n keys (dify-i18n/valid-i18n-keys rule)
+            - No extra keys (dify-i18n/no-extra-keys rule)
+
+            ### Step 3.2: Run Final i18n Check
+            ```bash
+            pnpm --dir ${{ github.workspace }}/web run i18n:check
+            ```
+
+            ### Step 3.3: Fix Any Remaining Issues
+            If check reports issues:
+            - Go back to PHASE 2 for unresolved items
+            - Repeat until check passes
+
+            ### Step 3.4: Generate Final Summary
+            ```
+            ╔══════════════════════════════════════════════════════════════╗
+            ║                    SYNC COMPLETED SUMMARY                    ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ Language      │ Added │ Updated │ Deleted │ Status          ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ zh-Hans       │  5    │   2     │    1    │ ✓ Complete      ║
+            ║ ja-JP         │  5    │   2     │    1    │ ✓ Complete      ║
+            ║ ...           │ ...   │  ...    │   ...   │ ...             ║
+            ╠══════════════════════════════════════════════════════════════╣
+            ║ i18n:check    │ PASSED - All keys in sync                   ║
+            ╚══════════════════════════════════════════════════════════════╝
+            ```
+
+            ## Mode-Specific Behavior
+
+            **SYNC_MODE = "incremental"** (default):
+            - Focus on keys identified from git diff
+            - Also check i18n:check output for any missing/extra keys
+            - Efficient for small changes
+
+            **SYNC_MODE = "full"**:
+            - Compare ALL keys between en-US and each language
+            - Run i18n:check to identify all discrepancies
+            - Use for first-time sync or fixing historical issues
+
+            ## Important Notes
+
+            1. Always run i18n:check BEFORE and AFTER making changes
+            2. The check script is the source of truth for missing/extra keys
+            3. For UPDATE scenario: git diff is the source of truth for changed values
+            4. Create a single commit with all translation changes
+            5. If any translation fails, continue with others and report failures
+
+            ═══════════════════════════════════════════════════════════════
+            ║  PHASE 4: COMMIT AND CREATE PR                              ║
+            ═══════════════════════════════════════════════════════════════
+
+            After all translations are complete and verified:
+
+            ### Step 4.1: Check for changes
+            ```bash
+            git -C ${{ github.workspace }} status --porcelain
+            ```
+
+            If there are changes:
+
+            ### Step 4.2: Create a new branch and commit
+            Run these git commands ONE BY ONE (not combined with &&).
+            **IMPORTANT**: Do NOT use `$()` command substitution. Use two separate commands:
+
+            1. First, get the timestamp:
+            ```bash
+            date +%Y%m%d-%H%M%S
+            ```
+            (Note the output, e.g., "20260115-143052")
+
+            2. Then create branch using the timestamp value:
+            ```bash
+            git -C ${{ github.workspace }} checkout -b chore/i18n-sync-20260115-143052
+            ```
+            (Replace "20260115-143052" with the actual timestamp from step 1)
+
+            3. Stage changes:
+            ```bash
+            git -C ${{ github.workspace }} add web/i18n/
+            ```
+
+            4. Commit:
+            ```bash
+            git -C ${{ github.workspace }} commit -m "chore(i18n): sync translations with en-US - Mode: ${{ steps.detect_changes.outputs.SYNC_MODE }}"
+            ```
+
+            5. Push:
+            ```bash
+            git -C ${{ github.workspace }} push origin HEAD
+            ```
+
+            ### Step 4.3: Create Pull Request
+            ```bash
+            gh pr create --repo ${{ github.repository }} --title "chore(i18n): sync translations with en-US" --body "## Summary
+
+            This PR was automatically generated to sync i18n translation files.
+
+            ### Changes
+            - Mode: ${{ steps.detect_changes.outputs.SYNC_MODE }}
+            - Files processed: ${{ steps.detect_changes.outputs.CHANGED_FILES }}
+
+            ### Verification
+            - [x] \`i18n:check\` passed
+            - [x] \`lint:fix\` applied
+
+            🤖 Generated with Claude Code GitHub Action" --base main
+            ```
+
+          claude_args: |
+            --max-turns 150
+            --allowedTools "Read,Write,Edit,Bash(git *),Bash(git:*),Bash(gh *),Bash(gh:*),Bash(pnpm *),Bash(pnpm:*),Bash(date *),Bash(date:*),Glob,Grep"
@@ -0,0 +1,66 @@
+name: Trigger i18n Sync on Push
+
+# This workflow bridges the push event to repository_dispatch
+# because claude-code-action doesn't support push events directly.
+# See: https://github.com/langgenius/dify/issues/30743
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'web/i18n/en-US/*.json'
+
+permissions:
+  contents: write
+
+jobs:
+  trigger:
+    if: github.repository == 'langgenius/dify'
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Detect changed files and generate diff
+        id: detect
+        run: |
+          BEFORE_SHA="${{ github.event.before }}"
+          # Handle edge case: force push may have null/zero SHA
+          if [ -z "$BEFORE_SHA" ] || [ "$BEFORE_SHA" = "0000000000000000000000000000000000000000" ]; then
+            BEFORE_SHA="HEAD~1"
+          fi
+
+          # Detect changed i18n files
+          changed=$(git diff --name-only "$BEFORE_SHA" "${{ github.sha }}" -- 'web/i18n/en-US/*.json' 2>/dev/null | xargs -n1 basename 2>/dev/null | sed 's/.json$//' | tr '\n' ' ' || echo "")
+          echo "changed_files=$changed" >> $GITHUB_OUTPUT
+
+          # Generate diff for context
+          git diff "$BEFORE_SHA" "${{ github.sha }}" -- 'web/i18n/en-US/*.json' > /tmp/i18n-diff.txt 2>/dev/null || echo "" > /tmp/i18n-diff.txt
+
+          # Truncate if too large (keep first 50KB to match receiving workflow)
+          head -c 50000 /tmp/i18n-diff.txt > /tmp/i18n-diff-truncated.txt
+          mv /tmp/i18n-diff-truncated.txt /tmp/i18n-diff.txt
+
+          # Base64 encode the diff for safe JSON transport (portable, single-line)
+          diff_base64=$(base64 < /tmp/i18n-diff.txt | tr -d '\n')
+          echo "diff_base64=$diff_base64" >> $GITHUB_OUTPUT
+
+          if [ -n "$changed" ]; then
+            echo "has_changes=true" >> $GITHUB_OUTPUT
+            echo "Detected changed files: $changed"
+          else
+            echo "has_changes=false" >> $GITHUB_OUTPUT
+            echo "No i18n changes detected"
+          fi
+
+      - name: Trigger i18n sync workflow
+        if: steps.detect.outputs.has_changes == 'true'
+        uses: peter-evans/repository-dispatch@v3
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          event-type: i18n-sync
+          client-payload: '{"changed_files": "${{ steps.detect.outputs.changed_files }}", "diff_base64": "${{ steps.detect.outputs.diff_base64 }}", "sync_mode": "incremental", "trigger_sha": "${{ github.sha }}"}'
@@ -31,7 +31,7 @@ jobs:
      - name: Setup Node.js
        uses: actions/setup-node@v6
        with:
-          node-version: 22
+          node-version: 24
          cache: pnpm
          cache-dependency-path: ./web/pnpm-lock.yaml

@@ -209,6 +209,7 @@ api/.vscode
 .history

 .idea/
+web/migration/

 # pnpm
 /.pnpm-store
@@ -1 +0,0 @@
-22.11.0
@@ -417,6 +417,8 @@ SMTP_USERNAME=123
 SMTP_PASSWORD=abc
 SMTP_USE_TLS=true
 SMTP_OPPORTUNISTIC_TLS=false
+# Optional: override the local hostname used for SMTP HELO/EHLO
+SMTP_LOCAL_HOSTNAME=
 # Sendgid configuration
 SENDGRID_API_KEY=
 # Sentry configuration
@@ -589,6 +591,7 @@ ENABLE_CLEAN_UNUSED_DATASETS_TASK=false
 ENABLE_CREATE_TIDB_SERVERLESS_TASK=false
 ENABLE_UPDATE_TIDB_SERVERLESS_STATUS_TASK=false
 ENABLE_CLEAN_MESSAGES=false
+ENABLE_WORKFLOW_RUN_CLEANUP_TASK=false
 ENABLE_MAIL_CLEAN_DOCUMENT_NOTIFY_TASK=false
 ENABLE_DATASETS_QUEUE_MONITOR=false
 ENABLE_CHECK_UPGRADABLE_PLUGIN_TASK=true
@@ -712,3 +715,4 @@ ANNOTATION_IMPORT_MAX_CONCURRENT=5
 SANDBOX_EXPIRED_RECORDS_CLEAN_GRACEFUL_PERIOD=21
 SANDBOX_EXPIRED_RECORDS_CLEAN_BATCH_SIZE=1000
 SANDBOX_EXPIRED_RECORDS_RETENTION_DAYS=30
+
@@ -1,7 +1,9 @@
 import base64
+import datetime
 import json
 import logging
 import secrets
+import time
 from typing import Any

 import click
@@ -34,7 +36,7 @@ from libs.rsa import generate_key_pair
 from models import Tenant
 from models.dataset import Dataset, DatasetCollectionBinding, DatasetMetadata, DatasetMetadataBinding, DocumentSegment
 from models.dataset import Document as DatasetDocument
-from models.model import Account, App, AppAnnotationSetting, AppMode, Conversation, MessageAnnotation, UploadFile
+from models.model import App, AppAnnotationSetting, AppMode, Conversation, MessageAnnotation, UploadFile
 from models.oauth import DatasourceOauthParamConfig, DatasourceProvider
 from models.provider import Provider, ProviderModel
 from models.provider_ids import DatasourceProviderID, ToolProviderID
@@ -45,6 +47,9 @@ from services.clear_free_plan_tenant_expired_logs import ClearFreePlanTenantExpi
 from services.plugin.data_migration import PluginDataMigration
 from services.plugin.plugin_migration import PluginMigration
 from services.plugin.plugin_service import PluginService
+from services.retention.conversation.messages_clean_policy import create_message_clean_policy
+from services.retention.conversation.messages_clean_service import MessagesCleanService
+from services.retention.workflow_run.clear_free_plan_expired_workflow_run_logs import WorkflowRunCleanup
 from tasks.remove_app_and_related_data_task import delete_draft_variables_batch

 logger = logging.getLogger(__name__)
@@ -62,8 +67,10 @@ def reset_password(email, new_password, password_confirm):
    if str(new_password).strip() != str(password_confirm).strip():
        click.echo(click.style("Passwords do not match.", fg="red"))
        return
+    normalized_email = email.strip().lower()
+
    with sessionmaker(db.engine, expire_on_commit=False).begin() as session:
-        account = session.query(Account).where(Account.email == email).one_or_none()
+        account = AccountService.get_account_by_email_with_case_fallback(email.strip(), session=session)

        if not account:
            click.echo(click.style(f"Account not found for email: {email}", fg="red"))
@@ -84,7 +91,7 @@ def reset_password(email, new_password, password_confirm):
        base64_password_hashed = base64.b64encode(password_hashed).decode()
        account.password = base64_password_hashed
        account.password_salt = base64_salt
-        AccountService.reset_login_error_rate_limit(email)
+        AccountService.reset_login_error_rate_limit(normalized_email)
        click.echo(click.style("Password reset successfully.", fg="green"))


@@ -100,20 +107,22 @@ def reset_email(email, new_email, email_confirm):
    if str(new_email).strip() != str(email_confirm).strip():
        click.echo(click.style("New emails do not match.", fg="red"))
        return
+    normalized_new_email = new_email.strip().lower()
+
    with sessionmaker(db.engine, expire_on_commit=False).begin() as session:
-        account = session.query(Account).where(Account.email == email).one_or_none()
+        account = AccountService.get_account_by_email_with_case_fallback(email.strip(), session=session)

        if not account:
            click.echo(click.style(f"Account not found for email: {email}", fg="red"))
            return

        try:
-            email_validate(new_email)
+            email_validate(normalized_new_email)
        except:
            click.echo(click.style(f"Invalid email: {new_email}", fg="red"))
            return

-        account.email = new_email
+        account.email = normalized_new_email
        click.echo(click.style("Email updated successfully.", fg="green"))


@@ -658,7 +667,7 @@ def create_tenant(email: str, language: str | None = None, name: str | None = No
        return

    # Create account
-    email = email.strip()
+    email = email.strip().lower()

    if "@" not in email:
        click.echo(click.style("Invalid email address.", fg="red"))
@@ -852,6 +861,61 @@ def clear_free_plan_tenant_expired_logs(days: int, batch: int, tenant_ids: list[
    click.echo(click.style("Clear free plan tenant expired logs completed.", fg="green"))


+@click.command("clean-workflow-runs", help="Clean expired workflow runs and related data for free tenants.")
+@click.option("--days", default=30, show_default=True, help="Delete workflow runs created before N days ago.")
+@click.option("--batch-size", default=200, show_default=True, help="Batch size for selecting workflow runs.")
+@click.option(
+    "--start-from",
+    type=click.DateTime(formats=["%Y-%m-%d", "%Y-%m-%dT%H:%M:%S"]),
+    default=None,
+    help="Optional lower bound (inclusive) for created_at; must be paired with --end-before.",
+)
+@click.option(
+    "--end-before",
+    type=click.DateTime(formats=["%Y-%m-%d", "%Y-%m-%dT%H:%M:%S"]),
+    default=None,
+    help="Optional upper bound (exclusive) for created_at; must be paired with --start-from.",
+)
+@click.option(
+    "--dry-run",
+    is_flag=True,
+    help="Preview cleanup results without deleting any workflow run data.",
+)
+def clean_workflow_runs(
+    days: int,
+    batch_size: int,
+    start_from: datetime.datetime | None,
+    end_before: datetime.datetime | None,
+    dry_run: bool,
+):
+    """
+    Clean workflow runs and related workflow data for free tenants.
+    """
+    if (start_from is None) ^ (end_before is None):
+        raise click.UsageError("--start-from and --end-before must be provided together.")
+
+    start_time = datetime.datetime.now(datetime.UTC)
+    click.echo(click.style(f"Starting workflow run cleanup at {start_time.isoformat()}.", fg="white"))
+
+    WorkflowRunCleanup(
+        days=days,
+        batch_size=batch_size,
+        start_from=start_from,
+        end_before=end_before,
+        dry_run=dry_run,
+    ).run()
+
+    end_time = datetime.datetime.now(datetime.UTC)
+    elapsed = end_time - start_time
+    click.echo(
+        click.style(
+            f"Workflow run cleanup completed. start={start_time.isoformat()} "
+            f"end={end_time.isoformat()} duration={elapsed}",
+            fg="green",
+        )
+    )
+
+
@click.option("-f", "--force", is_flag=True, help="Skip user confirmation and force the command to execute.")
@click.command("clear-orphaned-file-records", help="Clear orphaned file records.")
 def clear_orphaned_file_records(force: bool):
@@ -2111,3 +2175,79 @@ def migrate_oss(
            except Exception as e:
                db.session.rollback()
                click.echo(click.style(f"Failed to update DB storage_type: {str(e)}", fg="red"))
+
+
+@click.command("clean-expired-messages", help="Clean expired messages.")
+@click.option(
+    "--start-from",
+    type=click.DateTime(formats=["%Y-%m-%d", "%Y-%m-%dT%H:%M:%S"]),
+    required=True,
+    help="Lower bound (inclusive) for created_at.",
+)
+@click.option(
+    "--end-before",
+    type=click.DateTime(formats=["%Y-%m-%d", "%Y-%m-%dT%H:%M:%S"]),
+    required=True,
+    help="Upper bound (exclusive) for created_at.",
+)
+@click.option("--batch-size", default=1000, show_default=True, help="Batch size for selecting messages.")
+@click.option(
+    "--graceful-period",
+    default=21,
+    show_default=True,
+    help="Graceful period in days after subscription expiration, will be ignored when billing is disabled.",
+)
+@click.option("--dry-run", is_flag=True, default=False, help="Show messages logs would be cleaned without deleting")
+def clean_expired_messages(
+    batch_size: int,
+    graceful_period: int,
+    start_from: datetime.datetime,
+    end_before: datetime.datetime,
+    dry_run: bool,
+):
+    """
+    Clean expired messages and related data for tenants based on clean policy.
+    """
+    click.echo(click.style("clean_messages: start clean messages.", fg="green"))
+
+    start_at = time.perf_counter()
+
+    try:
+        # Create policy based on billing configuration
+        # NOTE: graceful_period will be ignored when billing is disabled.
+        policy = create_message_clean_policy(graceful_period_days=graceful_period)
+
+        # Create and run the cleanup service
+        service = MessagesCleanService.from_time_range(
+            policy=policy,
+            start_from=start_from,
+            end_before=end_before,
+            batch_size=batch_size,
+            dry_run=dry_run,
+        )
+        stats = service.run()
+
+        end_at = time.perf_counter()
+        click.echo(
+            click.style(
+                f"clean_messages: completed successfully\n"
+                f"  - Latency: {end_at - start_at:.2f}s\n"
+                f"  - Batches processed: {stats['batches']}\n"
+                f"  - Total messages scanned: {stats['total_messages']}\n"
+                f"  - Messages filtered: {stats['filtered_messages']}\n"
+                f"  - Messages deleted: {stats['total_deleted']}",
+                fg="green",
+            )
+        )
+    except Exception as e:
+        end_at = time.perf_counter()
+        logger.exception("clean_messages failed")
+        click.echo(
+            click.style(
+                f"clean_messages: failed after {end_at - start_at:.2f}s - {str(e)}",
+                fg="red",
+            )
+        )
+        raise
+
+    click.echo(click.style("messages cleanup completed.", fg="green"))
@@ -949,6 +949,12 @@ class MailConfig(BaseSettings):
        default=False,
    )

+    SMTP_LOCAL_HOSTNAME: str | None = Field(
+        description="Override the local hostname used in SMTP HELO/EHLO. "
+        "Useful behind NAT or when the default hostname causes rejections.",
+        default=None,
+    )
+
    EMAIL_SEND_IP_LIMIT_PER_MINUTE: PositiveInt = Field(
        description="Maximum number of emails allowed to be sent from the same IP address in a minute",
        default=50,
@@ -1101,6 +1107,10 @@ class CeleryScheduleTasksConfig(BaseSettings):
        description="Enable clean messages task",
        default=False,
    )
+    ENABLE_WORKFLOW_RUN_CLEANUP_TASK: bool = Field(
+        description="Enable scheduled workflow run cleanup task",
+        default=False,
+    )
    ENABLE_MAIL_CLEAN_DOCUMENT_NOTIFY_TASK: bool = Field(
        description="Enable mail clean document notify task",
        default=False,
@@ -8,6 +8,11 @@ class HostedCreditConfig(BaseSettings):
        default="",
    )

+    HOSTED_POOL_CREDITS: int = Field(
+        description="Pool credits for hosted service",
+        default=200,
+    )
+
    def get_model_credits(self, model_name: str) -> int:
        """
        Get credit value for a specific model name.
@@ -60,19 +65,46 @@ class HostedOpenAiConfig(BaseSettings):

    HOSTED_OPENAI_TRIAL_MODELS: str = Field(
        description="Comma-separated list of available models for trial access",
-        default="gpt-3.5-turbo,"
-        "gpt-3.5-turbo-1106,"
-        "gpt-3.5-turbo-instruct,"
+        default="gpt-4,"
+        "gpt-4-turbo-preview,"
+        "gpt-4-turbo-2024-04-09,"
+        "gpt-4-1106-preview,"
+        "gpt-4-0125-preview,"
+        "gpt-4-turbo,"
+        "gpt-4.1,"
+        "gpt-4.1-2025-04-14,"
+        "gpt-4.1-mini,"
+        "gpt-4.1-mini-2025-04-14,"
+        "gpt-4.1-nano,"
+        "gpt-4.1-nano-2025-04-14,"
+        "gpt-3.5-turbo,"
        "gpt-3.5-turbo-16k,"
        "gpt-3.5-turbo-16k-0613,"
+        "gpt-3.5-turbo-1106,"
        "gpt-3.5-turbo-0613,"
        "gpt-3.5-turbo-0125,"
-        "text-davinci-003",
-    )
-
-    HOSTED_OPENAI_QUOTA_LIMIT: NonNegativeInt = Field(
-        description="Quota limit for hosted OpenAI service usage",
-        default=200,
+        "gpt-3.5-turbo-instruct,"
+        "text-davinci-003,"
+        "chatgpt-4o-latest,"
+        "gpt-4o,"
+        "gpt-4o-2024-05-13,"
+        "gpt-4o-2024-08-06,"
+        "gpt-4o-2024-11-20,"
+        "gpt-4o-audio-preview,"
+        "gpt-4o-audio-preview-2025-06-03,"
+        "gpt-4o-mini,"
+        "gpt-4o-mini-2024-07-18,"
+        "o3-mini,"
+        "o3-mini-2025-01-31,"
+        "gpt-5-mini-2025-08-07,"
+        "gpt-5-mini,"
+        "o4-mini,"
+        "o4-mini-2025-04-16,"
+        "gpt-5-chat-latest,"
+        "gpt-5,"
+        "gpt-5-2025-08-07,"
+        "gpt-5-nano,"
+        "gpt-5-nano-2025-08-07",
    )

    HOSTED_OPENAI_PAID_ENABLED: bool = Field(
@@ -87,6 +119,13 @@ class HostedOpenAiConfig(BaseSettings):
        "gpt-4-turbo-2024-04-09,"
        "gpt-4-1106-preview,"
        "gpt-4-0125-preview,"
+        "gpt-4-turbo,"
+        "gpt-4.1,"
+        "gpt-4.1-2025-04-14,"
+        "gpt-4.1-mini,"
+        "gpt-4.1-mini-2025-04-14,"
+        "gpt-4.1-nano,"
+        "gpt-4.1-nano-2025-04-14,"
        "gpt-3.5-turbo,"
        "gpt-3.5-turbo-16k,"
        "gpt-3.5-turbo-16k-0613,"
@@ -94,7 +133,150 @@ class HostedOpenAiConfig(BaseSettings):
        "gpt-3.5-turbo-0613,"
        "gpt-3.5-turbo-0125,"
        "gpt-3.5-turbo-instruct,"
-        "text-davinci-003",
+        "text-davinci-003,"
+        "chatgpt-4o-latest,"
+        "gpt-4o,"
+        "gpt-4o-2024-05-13,"
+        "gpt-4o-2024-08-06,"
+        "gpt-4o-2024-11-20,"
+        "gpt-4o-audio-preview,"
+        "gpt-4o-audio-preview-2025-06-03,"
+        "gpt-4o-mini,"
+        "gpt-4o-mini-2024-07-18,"
+        "o3-mini,"
+        "o3-mini-2025-01-31,"
+        "gpt-5-mini-2025-08-07,"
+        "gpt-5-mini,"
+        "o4-mini,"
+        "o4-mini-2025-04-16,"
+        "gpt-5-chat-latest,"
+        "gpt-5,"
+        "gpt-5-2025-08-07,"
+        "gpt-5-nano,"
+        "gpt-5-nano-2025-08-07",
+    )
+
+
+class HostedGeminiConfig(BaseSettings):
+    """
+    Configuration for fetching Gemini service
+    """
+
+    HOSTED_GEMINI_API_KEY: str | None = Field(
+        description="API key for hosted Gemini service",
+        default=None,
+    )
+
+    HOSTED_GEMINI_API_BASE: str | None = Field(
+        description="Base URL for hosted Gemini API",
+        default=None,
+    )
+
+    HOSTED_GEMINI_API_ORGANIZATION: str | None = Field(
+        description="Organization ID for hosted Gemini service",
+        default=None,
+    )
+
+    HOSTED_GEMINI_TRIAL_ENABLED: bool = Field(
+        description="Enable trial access to hosted Gemini service",
+        default=False,
+    )
+
+    HOSTED_GEMINI_TRIAL_MODELS: str = Field(
+        description="Comma-separated list of available models for trial access",
+        default="gemini-2.5-flash,gemini-2.0-flash,gemini-2.0-flash-lite,",
+    )
+
+    HOSTED_GEMINI_PAID_ENABLED: bool = Field(
+        description="Enable paid access to hosted gemini service",
+        default=False,
+    )
+
+    HOSTED_GEMINI_PAID_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="gemini-2.5-flash,gemini-2.0-flash,gemini-2.0-flash-lite,",
+    )
+
+
+class HostedXAIConfig(BaseSettings):
+    """
+    Configuration for fetching XAI service
+    """
+
+    HOSTED_XAI_API_KEY: str | None = Field(
+        description="API key for hosted XAI service",
+        default=None,
+    )
+
+    HOSTED_XAI_API_BASE: str | None = Field(
+        description="Base URL for hosted XAI API",
+        default=None,
+    )
+
+    HOSTED_XAI_API_ORGANIZATION: str | None = Field(
+        description="Organization ID for hosted XAI service",
+        default=None,
+    )
+
+    HOSTED_XAI_TRIAL_ENABLED: bool = Field(
+        description="Enable trial access to hosted XAI service",
+        default=False,
+    )
+
+    HOSTED_XAI_TRIAL_MODELS: str = Field(
+        description="Comma-separated list of available models for trial access",
+        default="grok-3,grok-3-mini,grok-3-mini-fast",
+    )
+
+    HOSTED_XAI_PAID_ENABLED: bool = Field(
+        description="Enable paid access to hosted XAI service",
+        default=False,
+    )
+
+    HOSTED_XAI_PAID_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="grok-3,grok-3-mini,grok-3-mini-fast",
+    )
+
+
+class HostedDeepseekConfig(BaseSettings):
+    """
+    Configuration for fetching Deepseek service
+    """
+
+    HOSTED_DEEPSEEK_API_KEY: str | None = Field(
+        description="API key for hosted Deepseek service",
+        default=None,
+    )
+
+    HOSTED_DEEPSEEK_API_BASE: str | None = Field(
+        description="Base URL for hosted Deepseek API",
+        default=None,
+    )
+
+    HOSTED_DEEPSEEK_API_ORGANIZATION: str | None = Field(
+        description="Organization ID for hosted Deepseek service",
+        default=None,
+    )
+
+    HOSTED_DEEPSEEK_TRIAL_ENABLED: bool = Field(
+        description="Enable trial access to hosted Deepseek service",
+        default=False,
+    )
+
+    HOSTED_DEEPSEEK_TRIAL_MODELS: str = Field(
+        description="Comma-separated list of available models for trial access",
+        default="deepseek-chat,deepseek-reasoner",
+    )
+
+    HOSTED_DEEPSEEK_PAID_ENABLED: bool = Field(
+        description="Enable paid access to hosted Deepseek service",
+        default=False,
+    )
+
+    HOSTED_DEEPSEEK_PAID_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="deepseek-chat,deepseek-reasoner",
    )


@@ -144,16 +326,66 @@ class HostedAnthropicConfig(BaseSettings):
        default=False,
    )

-    HOSTED_ANTHROPIC_QUOTA_LIMIT: NonNegativeInt = Field(
-        description="Quota limit for hosted Anthropic service usage",
-        default=600000,
-    )
-
    HOSTED_ANTHROPIC_PAID_ENABLED: bool = Field(
        description="Enable paid access to hosted Anthropic service",
        default=False,
    )

+    HOSTED_ANTHROPIC_TRIAL_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="claude-opus-4-20250514,"
+        "claude-sonnet-4-20250514,"
+        "claude-3-5-haiku-20241022,"
+        "claude-3-opus-20240229,"
+        "claude-3-7-sonnet-20250219,"
+        "claude-3-haiku-20240307",
+    )
+    HOSTED_ANTHROPIC_PAID_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="claude-opus-4-20250514,"
+        "claude-sonnet-4-20250514,"
+        "claude-3-5-haiku-20241022,"
+        "claude-3-opus-20240229,"
+        "claude-3-7-sonnet-20250219,"
+        "claude-3-haiku-20240307",
+    )
+
+
+class HostedTongyiConfig(BaseSettings):
+    """
+    Configuration for hosted Tongyi service
+    """
+
+    HOSTED_TONGYI_API_KEY: str | None = Field(
+        description="API key for hosted Tongyi service",
+        default=None,
+    )
+
+    HOSTED_TONGYI_USE_INTERNATIONAL_ENDPOINT: bool = Field(
+        description="Use international endpoint for hosted Tongyi service",
+        default=False,
+    )
+
+    HOSTED_TONGYI_TRIAL_ENABLED: bool = Field(
+        description="Enable trial access to hosted Tongyi service",
+        default=False,
+    )
+
+    HOSTED_TONGYI_PAID_ENABLED: bool = Field(
+        description="Enable paid access to hosted Anthropic service",
+        default=False,
+    )
+
+    HOSTED_TONGYI_TRIAL_MODELS: str = Field(
+        description="Comma-separated list of available models for trial access",
+        default="",
+    )
+
+    HOSTED_TONGYI_PAID_MODELS: str = Field(
+        description="Comma-separated list of available models for paid access",
+        default="",
+    )
+

 class HostedMinmaxConfig(BaseSettings):
    """
@@ -246,9 +478,13 @@ class HostedServiceConfig(
    HostedOpenAiConfig,
    HostedSparkConfig,
    HostedZhipuAIConfig,
+    HostedTongyiConfig,
    # moderation
    HostedModerationConfig,
    # credit config
    HostedCreditConfig,
+    HostedGeminiConfig,
+    HostedXAIConfig,
+    HostedDeepseekConfig,
 ):
    pass
@@ -4,7 +4,7 @@ from pydantic_settings import BaseSettings

 class VolcengineTOSStorageConfig(BaseSettings):
    """
-    Configuration settings for Volcengine Tinder Object Storage (TOS)
+    Configuration settings for Volcengine Torch Object Storage (TOS)
    """

    VOLCENGINE_TOS_BUCKET_NAME: str | None = Field(
@@ -592,9 +592,12 @@ def _get_conversation(app_model, conversation_id):
    if not conversation:
        raise NotFound("Conversation Not Exists.")

-    if not conversation.read_at:
-        conversation.read_at = naive_utc_now()
-        conversation.read_account_id = current_user.id
-        db.session.commit()
+    db.session.execute(
+        sa.update(Conversation)
+        .where(Conversation.id == conversation_id, Conversation.read_at.is_(None))
+        .values(read_at=naive_utc_now(), read_account_id=current_user.id)
+    )
+    db.session.commit()
+    db.session.refresh(conversation)

    return conversation
@@ -55,6 +55,35 @@ class InstructionTemplatePayload(BaseModel):
    type: str = Field(..., description="Instruction template type")


+class ContextGeneratePayload(BaseModel):
+    """Payload for generating extractor code node."""
+
+    workflow_id: str = Field(..., description="Workflow ID")
+    node_id: str = Field(..., description="Current tool/llm node ID")
+    parameter_name: str = Field(..., description="Parameter name to generate code for")
+    language: str = Field(default="python3", description="Code language (python3/javascript)")
+    prompt_messages: list[dict[str, Any]] = Field(
+        ..., description="Multi-turn conversation history, last message is the current instruction"
+    )
+    model_config_data: dict[str, Any] = Field(..., alias="model_config", description="Model configuration")
+
+
+class SuggestedQuestionsPayload(BaseModel):
+    """Payload for generating suggested questions."""
+
+    workflow_id: str = Field(..., description="Workflow ID")
+    node_id: str = Field(..., description="Current tool/llm node ID")
+    parameter_name: str = Field(..., description="Parameter name")
+    language: str = Field(
+        default="English", description="Language for generated questions (e.g. English, Chinese, Japanese)"
+    )
+    model_config_data: dict[str, Any] | None = Field(
+        default=None,
+        alias="model_config",
+        description="Model configuration (optional, uses system default if not provided)",
+    )
+
+
 def reg(cls: type[BaseModel]):
    console_ns.schema_model(cls.__name__, cls.model_json_schema(ref_template=DEFAULT_REF_TEMPLATE_SWAGGER_2_0))

@@ -64,6 +93,8 @@ reg(RuleCodeGeneratePayload)
 reg(RuleStructuredOutputPayload)
 reg(InstructionGeneratePayload)
 reg(InstructionTemplatePayload)
+reg(ContextGeneratePayload)
+reg(SuggestedQuestionsPayload)


@console_ns.route("/rule-generate")
@@ -278,3 +309,74 @@ class InstructionGenerationTemplateApi(Resource):
                return {"data": INSTRUCTION_GENERATE_TEMPLATE_CODE}
            case _:
                raise ValueError(f"Invalid type: {args.type}")
+
+
+@console_ns.route("/context-generate")
+class ContextGenerateApi(Resource):
+    @console_ns.doc("generate_with_context")
+    @console_ns.doc(description="Generate with multi-turn conversation context")
+    @console_ns.expect(console_ns.models[ContextGeneratePayload.__name__])
+    @console_ns.response(200, "Content generated successfully")
+    @console_ns.response(400, "Invalid request parameters or workflow not found")
+    @console_ns.response(402, "Provider quota exceeded")
+    @setup_required
+    @login_required
+    @account_initialization_required
+    def post(self):
+        from core.llm_generator.utils import deserialize_prompt_messages
+
+        args = ContextGeneratePayload.model_validate(console_ns.payload)
+        _, current_tenant_id = current_account_with_tenant()
+
+        prompt_messages = deserialize_prompt_messages(args.prompt_messages)
+
+        try:
+            return LLMGenerator.generate_with_context(
+                tenant_id=current_tenant_id,
+                workflow_id=args.workflow_id,
+                node_id=args.node_id,
+                parameter_name=args.parameter_name,
+                language=args.language,
+                prompt_messages=prompt_messages,
+                model_config=args.model_config_data,
+            )
+        except ProviderTokenNotInitError as ex:
+            raise ProviderNotInitializeError(ex.description)
+        except QuotaExceededError:
+            raise ProviderQuotaExceededError()
+        except ModelCurrentlyNotSupportError:
+            raise ProviderModelCurrentlyNotSupportError()
+        except InvokeError as e:
+            raise CompletionRequestError(e.description)
+
+
+@console_ns.route("/context-generate/suggested-questions")
+class SuggestedQuestionsApi(Resource):
+    @console_ns.doc("generate_suggested_questions")
+    @console_ns.doc(description="Generate suggested questions for context generation")
+    @console_ns.expect(console_ns.models[SuggestedQuestionsPayload.__name__])
+    @console_ns.response(200, "Questions generated successfully")
+    @setup_required
+    @login_required
+    @account_initialization_required
+    def post(self):
+        args = SuggestedQuestionsPayload.model_validate(console_ns.payload)
+        _, current_tenant_id = current_account_with_tenant()
+
+        try:
+            return LLMGenerator.generate_suggested_questions(
+                tenant_id=current_tenant_id,
+                workflow_id=args.workflow_id,
+                node_id=args.node_id,
+                parameter_name=args.parameter_name,
+                language=args.language,
+                model_config=args.model_config_data,
+            )
+        except ProviderTokenNotInitError as ex:
+            raise ProviderNotInitializeError(ex.description)
+        except QuotaExceededError:
+            raise ProviderQuotaExceededError()
+        except ModelCurrentlyNotSupportError:
+            raise ProviderModelCurrentlyNotSupportError()
+        except InvokeError as e:
+            raise CompletionRequestError(e.description)
@@ -202,7 +202,6 @@ message_detail_model = console_ns.model(
        "status": fields.String,
        "error": fields.String,
        "parent_message_id": fields.String,
-        "generation_detail": fields.Raw,
    },
 )

@@ -63,10 +63,9 @@ class ActivateCheckApi(Resource):
        args = ActivateCheckQuery.model_validate(request.args.to_dict(flat=True))  # type: ignore

        workspaceId = args.workspace_id
-        reg_email = args.email
        token = args.token

-        invitation = RegisterService.get_invitation_if_token_valid(workspaceId, reg_email, token)
+        invitation = RegisterService.get_invitation_with_case_fallback(workspaceId, args.email, token)
        if invitation:
            data = invitation.get("data", {})
            tenant = invitation.get("tenant", None)
@@ -100,11 +99,12 @@ class ActivateApi(Resource):
    def post(self):
        args = ActivatePayload.model_validate(console_ns.payload)

-        invitation = RegisterService.get_invitation_if_token_valid(args.workspace_id, args.email, args.token)
+        normalized_request_email = args.email.lower() if args.email else None
+        invitation = RegisterService.get_invitation_with_case_fallback(args.workspace_id, args.email, args.token)
        if invitation is None:
            raise AlreadyActivateError()

-        RegisterService.revoke_token(args.workspace_id, args.email, args.token)
+        RegisterService.revoke_token(args.workspace_id, normalized_request_email, args.token)

        account = invitation["account"]
        account.name = args.name
@@ -1,7 +1,6 @@
 from flask import request
 from flask_restx import Resource
 from pydantic import BaseModel, Field, field_validator
-from sqlalchemy import select
 from sqlalchemy.orm import Session

 from configs import dify_config
@@ -62,6 +61,7 @@ class EmailRegisterSendEmailApi(Resource):
    @email_register_enabled
    def post(self):
        args = EmailRegisterSendPayload.model_validate(console_ns.payload)
+        normalized_email = args.email.lower()

        ip_address = extract_remote_ip(request)
        if AccountService.is_email_send_ip_limit(ip_address):
@@ -70,13 +70,12 @@ class EmailRegisterSendEmailApi(Resource):
        if args.language in languages:
            language = args.language

-        if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(args.email):
+        if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(normalized_email):
            raise AccountInFreezeError()

        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=args.email)).scalar_one_or_none()
-        token = None
-        token = AccountService.send_email_register_email(email=args.email, account=account, language=language)
+            account = AccountService.get_account_by_email_with_case_fallback(args.email, session=session)
+        token = AccountService.send_email_register_email(email=normalized_email, account=account, language=language)
        return {"result": "success", "data": token}


@@ -88,9 +87,9 @@ class EmailRegisterCheckApi(Resource):
    def post(self):
        args = EmailRegisterValidityPayload.model_validate(console_ns.payload)

-        user_email = args.email
+        user_email = args.email.lower()

-        is_email_register_error_rate_limit = AccountService.is_email_register_error_rate_limit(args.email)
+        is_email_register_error_rate_limit = AccountService.is_email_register_error_rate_limit(user_email)
        if is_email_register_error_rate_limit:
            raise EmailRegisterLimitError()

@@ -98,11 +97,14 @@ class EmailRegisterCheckApi(Resource):
        if token_data is None:
            raise InvalidTokenError()

-        if user_email != token_data.get("email"):
+        token_email = token_data.get("email")
+        normalized_token_email = token_email.lower() if isinstance(token_email, str) else token_email
+
+        if user_email != normalized_token_email:
            raise InvalidEmailError()

        if args.code != token_data.get("code"):
-            AccountService.add_email_register_error_rate_limit(args.email)
+            AccountService.add_email_register_error_rate_limit(user_email)
            raise EmailCodeError()

        # Verified, revoke the first token
@@ -113,8 +115,8 @@ class EmailRegisterCheckApi(Resource):
            user_email, code=args.code, additional_data={"phase": "register"}
        )

-        AccountService.reset_email_register_error_rate_limit(args.email)
-        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}
+        AccountService.reset_email_register_error_rate_limit(user_email)
+        return {"is_valid": True, "email": normalized_token_email, "token": new_token}


@console_ns.route("/email-register")
@@ -141,22 +143,23 @@ class EmailRegisterResetApi(Resource):
        AccountService.revoke_email_register_token(args.token)

        email = register_data.get("email", "")
+        normalized_email = email.lower()

        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(email, session=session)

            if account:
                raise EmailAlreadyInUseError()
            else:
-                account = self._create_new_account(email, args.password_confirm)
+                account = self._create_new_account(normalized_email, args.password_confirm)
                if not account:
                    raise AccountNotFoundError()
                token_pair = AccountService.login(account=account, ip_address=extract_remote_ip(request))
-                AccountService.reset_login_error_rate_limit(email)
+                AccountService.reset_login_error_rate_limit(normalized_email)

        return {"result": "success", "data": token_pair.model_dump()}

-    def _create_new_account(self, email, password) -> Account | None:
+    def _create_new_account(self, email: str, password: str) -> Account | None:
        # Create new account if allowed
        account = None
        try:
@@ -4,7 +4,6 @@ import secrets
 from flask import request
 from flask_restx import Resource, fields
 from pydantic import BaseModel, Field, field_validator
-from sqlalchemy import select
 from sqlalchemy.orm import Session

 from controllers.console import console_ns
@@ -21,7 +20,6 @@ from events.tenant_event import tenant_was_created
 from extensions.ext_database import db
 from libs.helper import EmailStr, extract_remote_ip
 from libs.password import hash_password, valid_password
-from models import Account
 from services.account_service import AccountService, TenantService
 from services.feature_service import FeatureService

@@ -76,6 +74,7 @@ class ForgotPasswordSendEmailApi(Resource):
    @email_password_login_enabled
    def post(self):
        args = ForgotPasswordSendPayload.model_validate(console_ns.payload)
+        normalized_email = args.email.lower()

        ip_address = extract_remote_ip(request)
        if AccountService.is_email_send_ip_limit(ip_address):
@@ -87,11 +86,11 @@ class ForgotPasswordSendEmailApi(Resource):
            language = "en-US"

        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=args.email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(args.email, session=session)

        token = AccountService.send_reset_password_email(
            account=account,
-            email=args.email,
+            email=normalized_email,
            language=language,
            is_allow_register=FeatureService.get_system_features().is_allow_register,
        )
@@ -122,9 +121,9 @@ class ForgotPasswordCheckApi(Resource):
    def post(self):
        args = ForgotPasswordCheckPayload.model_validate(console_ns.payload)

-        user_email = args.email
+        user_email = args.email.lower()

-        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(args.email)
+        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(user_email)
        if is_forgot_password_error_rate_limit:
            raise EmailPasswordResetLimitError()

@@ -132,11 +131,16 @@ class ForgotPasswordCheckApi(Resource):
        if token_data is None:
            raise InvalidTokenError()

-        if user_email != token_data.get("email"):
+        token_email = token_data.get("email")
+        if not isinstance(token_email, str):
+            raise InvalidEmailError()
+        normalized_token_email = token_email.lower()
+
+        if user_email != normalized_token_email:
            raise InvalidEmailError()

        if args.code != token_data.get("code"):
-            AccountService.add_forgot_password_error_rate_limit(args.email)
+            AccountService.add_forgot_password_error_rate_limit(user_email)
            raise EmailCodeError()

        # Verified, revoke the first token
@@ -144,11 +148,11 @@ class ForgotPasswordCheckApi(Resource):

        # Refresh token data by generating a new token
        _, new_token = AccountService.generate_reset_password_token(
-            user_email, code=args.code, additional_data={"phase": "reset"}
+            token_email, code=args.code, additional_data={"phase": "reset"}
        )

-        AccountService.reset_forgot_password_error_rate_limit(args.email)
-        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}
+        AccountService.reset_forgot_password_error_rate_limit(user_email)
+        return {"is_valid": True, "email": normalized_token_email, "token": new_token}


@console_ns.route("/forgot-password/resets")
@@ -187,9 +191,8 @@ class ForgotPasswordResetApi(Resource):
        password_hashed = hash_password(args.new_password, salt)

        email = reset_data.get("email", "")
-
        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(email, session=session)

            if account:
                self._update_existing_account(account, password_hashed, salt, session)
@@ -90,32 +90,38 @@ class LoginApi(Resource):
    def post(self):
        """Authenticate user and login."""
        args = LoginPayload.model_validate(console_ns.payload)
+        request_email = args.email
+        normalized_email = request_email.lower()

-        if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(args.email):
+        if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(normalized_email):
            raise AccountInFreezeError()

-        is_login_error_rate_limit = AccountService.is_login_error_rate_limit(args.email)
+        is_login_error_rate_limit = AccountService.is_login_error_rate_limit(normalized_email)
        if is_login_error_rate_limit:
            raise EmailPasswordLoginLimitError()

+        invite_token = args.invite_token
        invitation_data: dict[str, Any] | None = None
-        if args.invite_token:
-            invitation_data = RegisterService.get_invitation_if_token_valid(None, args.email, args.invite_token)
+        if invite_token:
+            invitation_data = RegisterService.get_invitation_with_case_fallback(None, request_email, invite_token)
+            if invitation_data is None:
+                invite_token = None

        try:
            if invitation_data:
                data = invitation_data.get("data", {})
                invitee_email = data.get("email") if data else None
-                if invitee_email != args.email:
+                invitee_email_normalized = invitee_email.lower() if isinstance(invitee_email, str) else invitee_email
+                if invitee_email_normalized != normalized_email:
                    raise InvalidEmailError()
-                account = AccountService.authenticate(args.email, args.password, args.invite_token)
-            else:
-                account = AccountService.authenticate(args.email, args.password)
+            account = _authenticate_account_with_case_fallback(
+                request_email, normalized_email, args.password, invite_token
+            )
        except services.errors.account.AccountLoginError:
            raise AccountBannedError()
-        except services.errors.account.AccountPasswordError:
-            AccountService.add_login_error_rate_limit(args.email)
-            raise AuthenticationFailedError()
+        except services.errors.account.AccountPasswordError as exc:
+            AccountService.add_login_error_rate_limit(normalized_email)
+            raise AuthenticationFailedError() from exc
        # SELF_HOSTED only have one workspace
        tenants = TenantService.get_join_tenants(account)
        if len(tenants) == 0:
@@ -130,7 +136,7 @@ class LoginApi(Resource):
                }

        token_pair = AccountService.login(account=account, ip_address=extract_remote_ip(request))
-        AccountService.reset_login_error_rate_limit(args.email)
+        AccountService.reset_login_error_rate_limit(normalized_email)

        # Create response with cookies instead of returning tokens in body
        response = make_response({"result": "success"})
@@ -170,18 +176,19 @@ class ResetPasswordSendEmailApi(Resource):
    @console_ns.expect(console_ns.models[EmailPayload.__name__])
    def post(self):
        args = EmailPayload.model_validate(console_ns.payload)
+        normalized_email = args.email.lower()

        if args.language is not None and args.language == "zh-Hans":
            language = "zh-Hans"
        else:
            language = "en-US"
        try:
-            account = AccountService.get_user_through_email(args.email)
+            account = _get_account_with_case_fallback(args.email)
        except AccountRegisterError:
            raise AccountInFreezeError()

        token = AccountService.send_reset_password_email(
-            email=args.email,
+            email=normalized_email,
            account=account,
            language=language,
            is_allow_register=FeatureService.get_system_features().is_allow_register,
@@ -196,6 +203,7 @@ class EmailCodeLoginSendEmailApi(Resource):
    @console_ns.expect(console_ns.models[EmailPayload.__name__])
    def post(self):
        args = EmailPayload.model_validate(console_ns.payload)
+        normalized_email = args.email.lower()

        ip_address = extract_remote_ip(request)
        if AccountService.is_email_send_ip_limit(ip_address):
@@ -206,13 +214,13 @@ class EmailCodeLoginSendEmailApi(Resource):
        else:
            language = "en-US"
        try:
-            account = AccountService.get_user_through_email(args.email)
+            account = _get_account_with_case_fallback(args.email)
        except AccountRegisterError:
            raise AccountInFreezeError()

        if account is None:
            if FeatureService.get_system_features().is_allow_register:
-                token = AccountService.send_email_code_login_email(email=args.email, language=language)
+                token = AccountService.send_email_code_login_email(email=normalized_email, language=language)
            else:
                raise AccountNotFound()
        else:
@@ -229,14 +237,17 @@ class EmailCodeLoginApi(Resource):
    def post(self):
        args = EmailCodeLoginPayload.model_validate(console_ns.payload)

-        user_email = args.email
+        original_email = args.email
+        user_email = original_email.lower()
        language = args.language

        token_data = AccountService.get_email_code_login_data(args.token)
        if token_data is None:
            raise InvalidTokenError()

-        if token_data["email"] != args.email:
+        token_email = token_data.get("email")
+        normalized_token_email = token_email.lower() if isinstance(token_email, str) else token_email
+        if normalized_token_email != user_email:
            raise InvalidEmailError()

        if token_data["code"] != args.code:
@@ -244,7 +255,7 @@ class EmailCodeLoginApi(Resource):

        AccountService.revoke_email_code_login_token(args.token)
        try:
-            account = AccountService.get_user_through_email(user_email)
+            account = _get_account_with_case_fallback(original_email)
        except AccountRegisterError:
            raise AccountInFreezeError()
        if account:
@@ -275,7 +286,7 @@ class EmailCodeLoginApi(Resource):
            except WorkspacesLimitExceededError:
                raise WorkspacesLimitExceeded()
        token_pair = AccountService.login(account, ip_address=extract_remote_ip(request))
-        AccountService.reset_login_error_rate_limit(args.email)
+        AccountService.reset_login_error_rate_limit(user_email)

        # Create response with cookies instead of returning tokens in body
        response = make_response({"result": "success"})
@@ -309,3 +320,22 @@ class RefreshTokenApi(Resource):
            return response
        except Exception as e:
            return {"result": "fail", "message": str(e)}, 401
+
+
+def _get_account_with_case_fallback(email: str):
+    account = AccountService.get_user_through_email(email)
+    if account or email == email.lower():
+        return account
+
+    return AccountService.get_user_through_email(email.lower())
+
+
+def _authenticate_account_with_case_fallback(
+    original_email: str, normalized_email: str, password: str, invite_token: str | None
+):
+    try:
+        return AccountService.authenticate(original_email, password, invite_token)
+    except services.errors.account.AccountPasswordError:
+        if original_email == normalized_email:
+            raise
+        return AccountService.authenticate(normalized_email, password, invite_token)
@@ -3,7 +3,6 @@ import logging
 import httpx
 from flask import current_app, redirect, request
 from flask_restx import Resource
-from sqlalchemy import select
 from sqlalchemy.orm import Session
 from werkzeug.exceptions import Unauthorized

@@ -118,7 +117,10 @@ class OAuthCallback(Resource):
            invitation = RegisterService.get_invitation_by_token(token=invite_token)
            if invitation:
                invitation_email = invitation.get("email", None)
-                if invitation_email != user_info.email:
+                invitation_email_normalized = (
+                    invitation_email.lower() if isinstance(invitation_email, str) else invitation_email
+                )
+                if invitation_email_normalized != user_info.email.lower():
                    return redirect(f"{dify_config.CONSOLE_WEB_URL}/signin?message=Invalid invitation token.")

            return redirect(f"{dify_config.CONSOLE_WEB_URL}/signin/invite-settings?invite_token={invite_token}")
@@ -175,7 +177,7 @@ def _get_account_by_openid_or_email(provider: str, user_info: OAuthUserInfo) ->

    if not account:
        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=user_info.email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(user_info.email, session=session)

    return account

@@ -197,9 +199,10 @@ def _generate_account(provider: str, user_info: OAuthUserInfo) -> tuple[Account,
                tenant_was_created.send(new_tenant)

    if not account:
+        normalized_email = user_info.email.lower()
        oauth_new_user = True
        if not FeatureService.get_system_features().is_allow_register:
-            if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(user_info.email):
+            if dify_config.BILLING_ENABLED and BillingService.is_email_in_freeze(normalized_email):
                raise AccountRegisterError(
                    description=(
                        "This email account has been deleted within the past "
@@ -210,7 +213,11 @@ def _generate_account(provider: str, user_info: OAuthUserInfo) -> tuple[Account,
                raise AccountRegisterError(description=("Invalid email or password"))
        account_name = user_info.name or "Dify"
        account = RegisterService.register(
-            email=user_info.email, name=account_name, password=None, open_id=user_info.id, provider=provider
+            email=normalized_email,
+            name=account_name,
+            password=None,
+            open_id=user_info.id,
+            provider=provider,
        )

        # Set interface language
@@ -7,7 +7,7 @@ from typing import Literal, cast
 import sqlalchemy as sa
 from flask import request
 from flask_restx import Resource, fields, marshal, marshal_with
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 from sqlalchemy import asc, desc, select
 from werkzeug.exceptions import Forbidden, NotFound

@@ -104,6 +104,15 @@ class DocumentRenamePayload(BaseModel):
    name: str


+class DocumentDatasetListParam(BaseModel):
+    page: int = Field(1, title="Page", description="Page number.")
+    limit: int = Field(20, title="Limit", description="Page size.")
+    search: str | None = Field(None, alias="keyword", title="Search", description="Search keyword.")
+    sort_by: str = Field("-created_at", alias="sort", title="SortBy", description="Sort by field.")
+    status: str | None = Field(None, title="Status", description="Document status.")
+    fetch_val: str = Field("false", alias="fetch")
+
+
 register_schema_models(
    console_ns,
    KnowledgeConfig,
@@ -225,14 +234,16 @@ class DatasetDocumentListApi(Resource):
    def get(self, dataset_id):
        current_user, current_tenant_id = current_account_with_tenant()
        dataset_id = str(dataset_id)
-        page = request.args.get("page", default=1, type=int)
-        limit = request.args.get("limit", default=20, type=int)
-        search = request.args.get("keyword", default=None, type=str)
-        sort = request.args.get("sort", default="-created_at", type=str)
-        status = request.args.get("status", default=None, type=str)
+        raw_args = request.args.to_dict()
+        param = DocumentDatasetListParam.model_validate(raw_args)
+        page = param.page
+        limit = param.limit
+        search = param.search
+        sort = param.sort_by
+        status = param.status
        # "yes", "true", "t", "y", "1" convert to True, while others convert to False.
        try:
-            fetch_val = request.args.get("fetch", default="false")
+            fetch_val = param.fetch_val
            if isinstance(fetch_val, bool):
                fetch = fetch_val
            else:
@@ -81,7 +81,7 @@ class ExternalKnowledgeApiPayload(BaseModel):
 class ExternalDatasetCreatePayload(BaseModel):
    external_knowledge_api_id: str
    external_knowledge_id: str
-    name: str = Field(..., min_length=1, max_length=40)
+    name: str = Field(..., min_length=1, max_length=100)
    description: str | None = Field(None, max_length=400)
    external_retrieval_model: dict[str, object] | None = None

@@ -84,10 +84,11 @@ class SetupApi(Resource):
            raise NotInitValidateError()

        args = SetupRequestPayload.model_validate(console_ns.payload)
+        normalized_email = args.email.lower()

        # setup
        RegisterService.setup(
-            email=args.email,
+            email=normalized_email,
            name=args.name,
            password=args.password,
            ip_address=extract_remote_ip(request),
@@ -41,7 +41,7 @@ from fields.member_fields import account_fields
 from libs.datetime_utils import naive_utc_now
 from libs.helper import EmailStr, TimestampField, extract_remote_ip, timezone
 from libs.login import current_account_with_tenant, login_required
-from models import Account, AccountIntegrate, InvitationCode
+from models import AccountIntegrate, InvitationCode
 from services.account_service import AccountService
 from services.billing_service import BillingService
 from services.errors.account import CurrentPasswordIncorrectError as ServiceCurrentPasswordIncorrectError
@@ -536,7 +536,8 @@ class ChangeEmailSendEmailApi(Resource):
        else:
            language = "en-US"
        account = None
-        user_email = args.email
+        user_email = None
+        email_for_sending = args.email.lower()
        if args.phase is not None and args.phase == "new_email":
            if args.token is None:
                raise InvalidTokenError()
@@ -546,16 +547,24 @@ class ChangeEmailSendEmailApi(Resource):
                raise InvalidTokenError()
            user_email = reset_data.get("email", "")

-            if user_email != current_user.email:
+            if user_email.lower() != current_user.email.lower():
                raise InvalidEmailError()
+
+            user_email = current_user.email
        else:
            with Session(db.engine) as session:
-                account = session.execute(select(Account).filter_by(email=args.email)).scalar_one_or_none()
+                account = AccountService.get_account_by_email_with_case_fallback(args.email, session=session)
            if account is None:
                raise AccountNotFound()
+            email_for_sending = account.email
+            user_email = account.email

        token = AccountService.send_change_email_email(
-            account=account, email=args.email, old_email=user_email, language=language, phase=args.phase
+            account=account,
+            email=email_for_sending,
+            old_email=user_email,
+            language=language,
+            phase=args.phase,
        )
        return {"result": "success", "data": token}

@@ -571,9 +580,9 @@ class ChangeEmailCheckApi(Resource):
        payload = console_ns.payload or {}
        args = ChangeEmailValidityPayload.model_validate(payload)

-        user_email = args.email
+        user_email = args.email.lower()

-        is_change_email_error_rate_limit = AccountService.is_change_email_error_rate_limit(args.email)
+        is_change_email_error_rate_limit = AccountService.is_change_email_error_rate_limit(user_email)
        if is_change_email_error_rate_limit:
            raise EmailChangeLimitError()

@@ -581,11 +590,13 @@ class ChangeEmailCheckApi(Resource):
        if token_data is None:
            raise InvalidTokenError()

-        if user_email != token_data.get("email"):
+        token_email = token_data.get("email")
+        normalized_token_email = token_email.lower() if isinstance(token_email, str) else token_email
+        if user_email != normalized_token_email:
            raise InvalidEmailError()

        if args.code != token_data.get("code"):
-            AccountService.add_change_email_error_rate_limit(args.email)
+            AccountService.add_change_email_error_rate_limit(user_email)
            raise EmailCodeError()

        # Verified, revoke the first token
@@ -596,8 +607,8 @@ class ChangeEmailCheckApi(Resource):
            user_email, code=args.code, old_email=token_data.get("old_email"), additional_data={}
        )

-        AccountService.reset_change_email_error_rate_limit(args.email)
-        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}
+        AccountService.reset_change_email_error_rate_limit(user_email)
+        return {"is_valid": True, "email": normalized_token_email, "token": new_token}


@console_ns.route("/account/change-email/reset")
@@ -611,11 +622,12 @@ class ChangeEmailResetApi(Resource):
    def post(self):
        payload = console_ns.payload or {}
        args = ChangeEmailResetPayload.model_validate(payload)
+        normalized_new_email = args.new_email.lower()

-        if AccountService.is_account_in_freeze(args.new_email):
+        if AccountService.is_account_in_freeze(normalized_new_email):
            raise AccountInFreezeError()

-        if not AccountService.check_email_unique(args.new_email):
+        if not AccountService.check_email_unique(normalized_new_email):
            raise EmailAlreadyInUseError()

        reset_data = AccountService.get_change_email_data(args.token)
@@ -626,13 +638,13 @@ class ChangeEmailResetApi(Resource):

        old_email = reset_data.get("old_email", "")
        current_user, _ = current_account_with_tenant()
-        if current_user.email != old_email:
+        if current_user.email.lower() != old_email.lower():
            raise AccountNotFound()

-        updated_account = AccountService.update_account_email(current_user, email=args.new_email)
+        updated_account = AccountService.update_account_email(current_user, email=normalized_new_email)

        AccountService.send_change_email_completed_notify_email(
-            email=args.new_email,
+            email=normalized_new_email,
        )

        return updated_account
@@ -645,8 +657,9 @@ class CheckEmailUnique(Resource):
    def post(self):
        payload = console_ns.payload or {}
        args = CheckEmailUniquePayload.model_validate(payload)
-        if AccountService.is_account_in_freeze(args.email):
+        normalized_email = args.email.lower()
+        if AccountService.is_account_in_freeze(normalized_email):
            raise AccountInFreezeError()
-        if not AccountService.check_email_unique(args.email):
+        if not AccountService.check_email_unique(normalized_email):
            raise EmailAlreadyInUseError()
        return {"result": "success"}
@@ -116,26 +116,31 @@ class MemberInviteEmailApi(Resource):
            raise WorkspaceMembersLimitExceeded()

        for invitee_email in invitee_emails:
+            normalized_invitee_email = invitee_email.lower()
            try:
                if not inviter.current_tenant:
                    raise ValueError("No current tenant")
                token = RegisterService.invite_new_member(
-                    inviter.current_tenant, invitee_email, interface_language, role=invitee_role, inviter=inviter
+                    tenant=inviter.current_tenant,
+                    email=invitee_email,
+                    language=interface_language,
+                    role=invitee_role,
+                    inviter=inviter,
                )
-                encoded_invitee_email = parse.quote(invitee_email)
+                encoded_invitee_email = parse.quote(normalized_invitee_email)
                invitation_results.append(
                    {
                        "status": "success",
-                        "email": invitee_email,
+                        "email": normalized_invitee_email,
                        "url": f"{console_web_url}/activate?email={encoded_invitee_email}&token={token}",
                    }
                )
            except AccountAlreadyInTenantError:
                invitation_results.append(
-                    {"status": "success", "email": invitee_email, "url": f"{console_web_url}/signin"}
+                    {"status": "success", "email": normalized_invitee_email, "url": f"{console_web_url}/signin"}
                )
            except Exception as e:
-                invitation_results.append({"status": "failed", "email": invitee_email, "message": str(e)})
+                invitation_results.append({"status": "failed", "email": normalized_invitee_email, "message": str(e)})

        return {
            "result": "success",
@@ -1,14 +1,14 @@
 import logging
-from collections.abc import Mapping
 from typing import Any

 from flask import make_response, redirect, request
-from flask_restx import Resource, reqparse
-from pydantic import BaseModel, Field, model_validator
+from flask_restx import Resource
+from pydantic import BaseModel, model_validator
 from sqlalchemy.orm import Session
 from werkzeug.exceptions import BadRequest, Forbidden

 from configs import dify_config
+from controllers.common.schema import register_schema_models
 from controllers.web.error import NotFoundError
 from core.model_runtime.utils.encoders import jsonable_encoder
 from core.plugin.entities.plugin_daemon import CredentialType
@@ -35,35 +35,38 @@ from ..wraps import (
 logger = logging.getLogger(__name__)


-class TriggerSubscriptionUpdateRequest(BaseModel):
-    """Request payload for updating a trigger subscription"""
+class TriggerSubscriptionBuilderCreatePayload(BaseModel):
+    credential_type: str = CredentialType.UNAUTHORIZED

-    name: str | None = Field(default=None, description="The name for the subscription")
-    credentials: Mapping[str, Any] | None = Field(default=None, description="The credentials for the subscription")
-    parameters: Mapping[str, Any] | None = Field(default=None, description="The parameters for the subscription")
-    properties: Mapping[str, Any] | None = Field(default=None, description="The properties for the subscription")
+
+class TriggerSubscriptionBuilderVerifyPayload(BaseModel):
+    credentials: dict[str, Any]
+
+
+class TriggerSubscriptionBuilderUpdatePayload(BaseModel):
+    name: str | None = None
+    parameters: dict[str, Any] | None = None
+    properties: dict[str, Any] | None = None
+    credentials: dict[str, Any] | None = None

    @model_validator(mode="after")
    def check_at_least_one_field(self):
-        if all(v is None for v in (self.name, self.credentials, self.parameters, self.properties)):
+        if all(v is None for v in self.model_dump().values()):
            raise ValueError("At least one of name, credentials, parameters, or properties must be provided")
        return self


-class TriggerSubscriptionVerifyRequest(BaseModel):
-    """Request payload for verifying subscription credentials."""
-
-    credentials: Mapping[str, Any] = Field(description="The credentials to verify")
+class TriggerOAuthClientPayload(BaseModel):
+    client_params: dict[str, Any] | None = None
+    enabled: bool | None = None


-console_ns.schema_model(
-    TriggerSubscriptionUpdateRequest.__name__,
-    TriggerSubscriptionUpdateRequest.model_json_schema(ref_template="#/definitions/{model}"),
-)
-
-console_ns.schema_model(
-    TriggerSubscriptionVerifyRequest.__name__,
-    TriggerSubscriptionVerifyRequest.model_json_schema(ref_template="#/definitions/{model}"),
+register_schema_models(
+    console_ns,
+    TriggerSubscriptionBuilderCreatePayload,
+    TriggerSubscriptionBuilderVerifyPayload,
+    TriggerSubscriptionBuilderUpdatePayload,
+    TriggerOAuthClientPayload,
 )


@@ -132,16 +135,11 @@ class TriggerSubscriptionListApi(Resource):
            raise


-parser = reqparse.RequestParser().add_argument(
-    "credential_type", type=str, required=False, nullable=True, location="json"
-)
-
-
@console_ns.route(
    "/workspaces/current/trigger-provider/<path:provider>/subscriptions/builder/create",
 )
 class TriggerSubscriptionBuilderCreateApi(Resource):
-    @console_ns.expect(parser)
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderCreatePayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -151,10 +149,10 @@ class TriggerSubscriptionBuilderCreateApi(Resource):
        user = current_user
        assert user.current_tenant_id is not None

-        args = parser.parse_args()
+        payload = TriggerSubscriptionBuilderCreatePayload.model_validate(console_ns.payload or {})

        try:
-            credential_type = CredentialType.of(args.get("credential_type") or CredentialType.UNAUTHORIZED.value)
+            credential_type = CredentialType.of(payload.credential_type)
            subscription_builder = TriggerSubscriptionBuilderService.create_trigger_subscription_builder(
                tenant_id=user.current_tenant_id,
                user_id=user.id,
@@ -182,18 +180,11 @@ class TriggerSubscriptionBuilderGetApi(Resource):
        )


-parser_api = (
-    reqparse.RequestParser()
-    # The credentials of the subscription builder
-    .add_argument("credentials", type=dict, required=False, nullable=True, location="json")
-)
-
-
@console_ns.route(
    "/workspaces/current/trigger-provider/<path:provider>/subscriptions/builder/verify-and-update/<path:subscription_builder_id>",
 )
-class TriggerSubscriptionBuilderVerifyAndUpdateApi(Resource):
-    @console_ns.expect(parser_api)
+class TriggerSubscriptionBuilderVerifyApi(Resource):
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderVerifyPayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -203,7 +194,7 @@ class TriggerSubscriptionBuilderVerifyAndUpdateApi(Resource):
        user = current_user
        assert user.current_tenant_id is not None

-        args = parser_api.parse_args()
+        payload = TriggerSubscriptionBuilderVerifyPayload.model_validate(console_ns.payload or {})

        try:
            # Use atomic update_and_verify to prevent race conditions
@@ -213,7 +204,7 @@ class TriggerSubscriptionBuilderVerifyAndUpdateApi(Resource):
                provider_id=TriggerProviderID(provider),
                subscription_builder_id=subscription_builder_id,
                subscription_builder_updater=SubscriptionBuilderUpdater(
-                    credentials=args.get("credentials", None),
+                    credentials=payload.credentials,
                ),
            )
        except Exception as e:
@@ -221,24 +212,11 @@ class TriggerSubscriptionBuilderVerifyAndUpdateApi(Resource):
            raise ValueError(str(e)) from e


-parser_update_api = (
-    reqparse.RequestParser()
-    # The name of the subscription builder
-    .add_argument("name", type=str, required=False, nullable=True, location="json")
-    # The parameters of the subscription builder
-    .add_argument("parameters", type=dict, required=False, nullable=True, location="json")
-    # The properties of the subscription builder
-    .add_argument("properties", type=dict, required=False, nullable=True, location="json")
-    # The credentials of the subscription builder
-    .add_argument("credentials", type=dict, required=False, nullable=True, location="json")
-)
-
-
@console_ns.route(
    "/workspaces/current/trigger-provider/<path:provider>/subscriptions/builder/update/<path:subscription_builder_id>",
 )
 class TriggerSubscriptionBuilderUpdateApi(Resource):
-    @console_ns.expect(parser_update_api)
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderUpdatePayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -249,7 +227,7 @@ class TriggerSubscriptionBuilderUpdateApi(Resource):
        assert isinstance(user, Account)
        assert user.current_tenant_id is not None

-        args = parser_update_api.parse_args()
+        payload = TriggerSubscriptionBuilderUpdatePayload.model_validate(console_ns.payload or {})
        try:
            return jsonable_encoder(
                TriggerSubscriptionBuilderService.update_trigger_subscription_builder(
@@ -257,10 +235,10 @@ class TriggerSubscriptionBuilderUpdateApi(Resource):
                    provider_id=TriggerProviderID(provider),
                    subscription_builder_id=subscription_builder_id,
                    subscription_builder_updater=SubscriptionBuilderUpdater(
-                        name=args.get("name", None),
-                        parameters=args.get("parameters", None),
-                        properties=args.get("properties", None),
-                        credentials=args.get("credentials", None),
+                        name=payload.name,
+                        parameters=payload.parameters,
+                        properties=payload.properties,
+                        credentials=payload.credentials,
                    ),
                )
            )
@@ -295,7 +273,7 @@ class TriggerSubscriptionBuilderLogsApi(Resource):
    "/workspaces/current/trigger-provider/<path:provider>/subscriptions/builder/build/<path:subscription_builder_id>",
 )
 class TriggerSubscriptionBuilderBuildApi(Resource):
-    @console_ns.expect(parser_update_api)
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderUpdatePayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -304,7 +282,7 @@ class TriggerSubscriptionBuilderBuildApi(Resource):
        """Build a subscription instance for a trigger provider"""
        user = current_user
        assert user.current_tenant_id is not None
-        args = parser_update_api.parse_args()
+        payload = TriggerSubscriptionBuilderUpdatePayload.model_validate(console_ns.payload or {})
        try:
            # Use atomic update_and_build to prevent race conditions
            TriggerSubscriptionBuilderService.update_and_build_builder(
@@ -313,9 +291,9 @@ class TriggerSubscriptionBuilderBuildApi(Resource):
                provider_id=TriggerProviderID(provider),
                subscription_builder_id=subscription_builder_id,
                subscription_builder_updater=SubscriptionBuilderUpdater(
-                    name=args.get("name", None),
-                    parameters=args.get("parameters", None),
-                    properties=args.get("properties", None),
+                    name=payload.name,
+                    parameters=payload.parameters,
+                    properties=payload.properties,
                ),
            )
            return 200
@@ -328,7 +306,7 @@ class TriggerSubscriptionBuilderBuildApi(Resource):
    "/workspaces/current/trigger-provider/<path:subscription_id>/subscriptions/update",
 )
 class TriggerSubscriptionUpdateApi(Resource):
-    @console_ns.expect(console_ns.models[TriggerSubscriptionUpdateRequest.__name__])
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderUpdatePayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -338,7 +316,7 @@ class TriggerSubscriptionUpdateApi(Resource):
        user = current_user
        assert user.current_tenant_id is not None

-        request = TriggerSubscriptionUpdateRequest.model_validate(console_ns.payload)
+        request = TriggerSubscriptionBuilderUpdatePayload.model_validate(console_ns.payload or {})

        subscription = TriggerProviderService.get_subscription_by_id(
            tenant_id=user.current_tenant_id,
@@ -568,13 +546,6 @@ class TriggerOAuthCallbackApi(Resource):
        return redirect(f"{dify_config.CONSOLE_WEB_URL}/oauth-callback")


-parser_oauth_client = (
-    reqparse.RequestParser()
-    .add_argument("client_params", type=dict, required=False, nullable=True, location="json")
-    .add_argument("enabled", type=bool, required=False, nullable=True, location="json")
-)
-
-
@console_ns.route("/workspaces/current/trigger-provider/<path:provider>/oauth/client")
 class TriggerOAuthClientManageApi(Resource):
    @setup_required
@@ -622,7 +593,7 @@ class TriggerOAuthClientManageApi(Resource):
            logger.exception("Error getting OAuth client", exc_info=e)
            raise

-    @console_ns.expect(parser_oauth_client)
+    @console_ns.expect(console_ns.models[TriggerOAuthClientPayload.__name__])
    @setup_required
    @login_required
    @is_admin_or_owner_required
@@ -632,15 +603,15 @@ class TriggerOAuthClientManageApi(Resource):
        user = current_user
        assert user.current_tenant_id is not None

-        args = parser_oauth_client.parse_args()
+        payload = TriggerOAuthClientPayload.model_validate(console_ns.payload or {})

        try:
            provider_id = TriggerProviderID(provider)
            return TriggerProviderService.save_custom_oauth_client_params(
                tenant_id=user.current_tenant_id,
                provider_id=provider_id,
-                client_params=args.get("client_params"),
-                enabled=args.get("enabled"),
+                client_params=payload.client_params,
+                enabled=payload.enabled,
            )

        except ValueError as e:
@@ -676,7 +647,7 @@ class TriggerOAuthClientManageApi(Resource):
    "/workspaces/current/trigger-provider/<path:provider>/subscriptions/verify/<path:subscription_id>",
 )
 class TriggerSubscriptionVerifyApi(Resource):
-    @console_ns.expect(console_ns.models[TriggerSubscriptionVerifyRequest.__name__])
+    @console_ns.expect(console_ns.models[TriggerSubscriptionBuilderVerifyPayload.__name__])
    @setup_required
    @login_required
    @edit_permission_required
@@ -686,9 +657,7 @@ class TriggerSubscriptionVerifyApi(Resource):
        user = current_user
        assert user.current_tenant_id is not None

-        verify_request: TriggerSubscriptionVerifyRequest = TriggerSubscriptionVerifyRequest.model_validate(
-            console_ns.payload
-        )
+        verify_request = TriggerSubscriptionBuilderVerifyPayload.model_validate(console_ns.payload or {})

        try:
            result = TriggerProviderService.verify_subscription_credentials(
@@ -80,6 +80,9 @@ tenant_fields = {
    "in_trial": fields.Boolean,
    "trial_end_reason": fields.String,
    "custom_config": fields.Raw(attribute="custom_config"),
+    "trial_credits": fields.Integer,
+    "trial_credits_used": fields.Integer,
+    "next_credit_reset_date": fields.Integer,
 }

 tenants_fields = {
@@ -4,7 +4,6 @@ import secrets
 from flask import request
 from flask_restx import Resource
 from pydantic import BaseModel, Field, field_validator
-from sqlalchemy import select
 from sqlalchemy.orm import Session

 from controllers.common.schema import register_schema_models
@@ -22,7 +21,7 @@ from controllers.web import web_ns
 from extensions.ext_database import db
 from libs.helper import EmailStr, extract_remote_ip
 from libs.password import hash_password, valid_password
-from models import Account
+from models.account import Account
 from services.account_service import AccountService


@@ -70,6 +69,9 @@ class ForgotPasswordSendEmailApi(Resource):
    def post(self):
        payload = ForgotPasswordSendPayload.model_validate(web_ns.payload or {})

+        request_email = payload.email
+        normalized_email = request_email.lower()
+
        ip_address = extract_remote_ip(request)
        if AccountService.is_email_send_ip_limit(ip_address):
            raise EmailSendIpLimitError()
@@ -80,12 +82,12 @@ class ForgotPasswordSendEmailApi(Resource):
            language = "en-US"

        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=payload.email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(request_email, session=session)
        token = None
        if account is None:
            raise AuthenticationFailedError()
        else:
-            token = AccountService.send_reset_password_email(account=account, email=payload.email, language=language)
+            token = AccountService.send_reset_password_email(account=account, email=normalized_email, language=language)

        return {"result": "success", "data": token}

@@ -104,9 +106,9 @@ class ForgotPasswordCheckApi(Resource):
    def post(self):
        payload = ForgotPasswordCheckPayload.model_validate(web_ns.payload or {})

-        user_email = payload.email
+        user_email = payload.email.lower()

-        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(payload.email)
+        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(user_email)
        if is_forgot_password_error_rate_limit:
            raise EmailPasswordResetLimitError()

@@ -114,11 +116,16 @@ class ForgotPasswordCheckApi(Resource):
        if token_data is None:
            raise InvalidTokenError()

-        if user_email != token_data.get("email"):
+        token_email = token_data.get("email")
+        if not isinstance(token_email, str):
+            raise InvalidEmailError()
+        normalized_token_email = token_email.lower()
+
+        if user_email != normalized_token_email:
            raise InvalidEmailError()

        if payload.code != token_data.get("code"):
-            AccountService.add_forgot_password_error_rate_limit(payload.email)
+            AccountService.add_forgot_password_error_rate_limit(user_email)
            raise EmailCodeError()

        # Verified, revoke the first token
@@ -126,11 +133,11 @@ class ForgotPasswordCheckApi(Resource):

        # Refresh token data by generating a new token
        _, new_token = AccountService.generate_reset_password_token(
-            user_email, code=payload.code, additional_data={"phase": "reset"}
+            token_email, code=payload.code, additional_data={"phase": "reset"}
        )

-        AccountService.reset_forgot_password_error_rate_limit(payload.email)
-        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}
+        AccountService.reset_forgot_password_error_rate_limit(user_email)
+        return {"is_valid": True, "email": normalized_token_email, "token": new_token}


@web_ns.route("/forgot-password/resets")
@@ -174,7 +181,7 @@ class ForgotPasswordResetApi(Resource):
        email = reset_data.get("email", "")

        with Session(db.engine) as session:
-            account = session.execute(select(Account).filter_by(email=email)).scalar_one_or_none()
+            account = AccountService.get_account_by_email_with_case_fallback(email, session=session)

            if account:
                self._update_existing_account(account, password_hashed, salt, session)
@@ -10,7 +10,12 @@ from controllers.console.auth.error import (
    InvalidEmailError,
 )
 from controllers.console.error import AccountBannedError
-from controllers.console.wraps import only_edition_enterprise, setup_required
+from controllers.console.wraps import (
+    decrypt_code_field,
+    decrypt_password_field,
+    only_edition_enterprise,
+    setup_required,
+)
 from controllers.web import web_ns
 from controllers.web.wraps import decode_jwt_token
 from libs.helper import email
@@ -42,6 +47,7 @@ class LoginApi(Resource):
            404: "Account not found",
        }
    )
+    @decrypt_password_field
    def post(self):
        """Authenticate user and login."""
        parser = (
@@ -181,6 +187,7 @@ class EmailCodeLoginApi(Resource):
            404: "Account not found",
        }
    )
+    @decrypt_code_field
    def post(self):
        parser = (
            reqparse.RequestParser()
@@ -190,25 +197,29 @@ class EmailCodeLoginApi(Resource):
        )
        args = parser.parse_args()

-        user_email = args["email"]
+        user_email = args["email"].lower()

        token_data = WebAppAuthService.get_email_code_login_data(args["token"])
        if token_data is None:
            raise InvalidTokenError()

-        if token_data["email"] != args["email"]:
+        token_email = token_data.get("email")
+        if not isinstance(token_email, str):
+            raise InvalidEmailError()
+        normalized_token_email = token_email.lower()
+        if normalized_token_email != user_email:
            raise InvalidEmailError()

        if token_data["code"] != args["code"]:
            raise EmailCodeError()

        WebAppAuthService.revoke_email_code_login_token(args["token"])
-        account = WebAppAuthService.get_user_through_email(user_email)
+        account = WebAppAuthService.get_user_through_email(token_email)
        if not account:
            raise AuthenticationFailedError()

        token = WebAppAuthService.login(account=account)
-        AccountService.reset_login_error_rate_limit(args["email"])
+        AccountService.reset_login_error_rate_limit(user_email)
        response = make_response({"result": "success", "data": {"access_token": token}})
        # set_access_token_to_cookie(request, response, token, samesite="None", httponly=False)
        return response
@@ -1,380 +0,0 @@
-import logging
-from collections.abc import Generator
-from copy import deepcopy
-from typing import Any
-
-from core.agent.base_agent_runner import BaseAgentRunner
-from core.agent.entities import AgentEntity, AgentLog, AgentResult
-from core.agent.patterns.strategy_factory import StrategyFactory
-from core.app.apps.base_app_queue_manager import PublishFrom
-from core.app.entities.queue_entities import QueueAgentThoughtEvent, QueueMessageEndEvent, QueueMessageFileEvent
-from core.file import file_manager
-from core.model_runtime.entities import (
-    AssistantPromptMessage,
-    LLMResult,
-    LLMResultChunk,
-    LLMUsage,
-    PromptMessage,
-    PromptMessageContentType,
-    SystemPromptMessage,
-    TextPromptMessageContent,
-    UserPromptMessage,
-)
-from core.model_runtime.entities.message_entities import ImagePromptMessageContent, PromptMessageContentUnionTypes
-from core.prompt.agent_history_prompt_transform import AgentHistoryPromptTransform
-from core.tools.__base.tool import Tool
-from core.tools.entities.tool_entities import ToolInvokeMeta
-from core.tools.tool_engine import ToolEngine
-from models.model import Message
-
-logger = logging.getLogger(__name__)
-
-
-class AgentAppRunner(BaseAgentRunner):
-    def _create_tool_invoke_hook(self, message: Message):
-        """
-        Create a tool invoke hook that uses ToolEngine.agent_invoke.
-        This hook handles file creation and returns proper meta information.
-        """
-        # Get trace manager from app generate entity
-        trace_manager = self.application_generate_entity.trace_manager
-
-        def tool_invoke_hook(
-            tool: Tool, tool_args: dict[str, Any], tool_name: str
-        ) -> tuple[str, list[str], ToolInvokeMeta]:
-            """Hook that uses agent_invoke for proper file and meta handling."""
-            tool_invoke_response, message_files, tool_invoke_meta = ToolEngine.agent_invoke(
-                tool=tool,
-                tool_parameters=tool_args,
-                user_id=self.user_id,
-                tenant_id=self.tenant_id,
-                message=message,
-                invoke_from=self.application_generate_entity.invoke_from,
-                agent_tool_callback=self.agent_callback,
-                trace_manager=trace_manager,
-                app_id=self.application_generate_entity.app_config.app_id,
-                message_id=message.id,
-                conversation_id=self.conversation.id,
-            )
-
-            # Publish files and track IDs
-            for message_file_id in message_files:
-                self.queue_manager.publish(
-                    QueueMessageFileEvent(message_file_id=message_file_id),
-                    PublishFrom.APPLICATION_MANAGER,
-                )
-                self._current_message_file_ids.append(message_file_id)
-
-            return tool_invoke_response, message_files, tool_invoke_meta
-
-        return tool_invoke_hook
-
-    def run(self, message: Message, query: str, **kwargs: Any) -> Generator[LLMResultChunk, None, None]:
-        """
-        Run Agent application
-        """
-        self.query = query
-        app_generate_entity = self.application_generate_entity
-
-        app_config = self.app_config
-        assert app_config is not None, "app_config is required"
-        assert app_config.agent is not None, "app_config.agent is required"
-
-        # convert tools into ModelRuntime Tool format
-        tool_instances, _ = self._init_prompt_tools()
-
-        assert app_config.agent
-
-        # Create tool invoke hook for agent_invoke
-        tool_invoke_hook = self._create_tool_invoke_hook(message)
-
-        # Get instruction for ReAct strategy
-        instruction = self.app_config.prompt_template.simple_prompt_template or ""
-
-        # Use factory to create appropriate strategy
-        strategy = StrategyFactory.create_strategy(
-            model_features=self.model_features,
-            model_instance=self.model_instance,
-            tools=list(tool_instances.values()),
-            files=list(self.files),
-            max_iterations=app_config.agent.max_iteration,
-            context=self.build_execution_context(),
-            agent_strategy=self.config.strategy,
-            tool_invoke_hook=tool_invoke_hook,
-            instruction=instruction,
-        )
-
-        # Initialize state variables
-        current_agent_thought_id = None
-        has_published_thought = False
-        current_tool_name: str | None = None
-        self._current_message_file_ids: list[str] = []
-
-        # organize prompt messages
-        prompt_messages = self._organize_prompt_messages()
-
-        # Run strategy
-        generator = strategy.run(
-            prompt_messages=prompt_messages,
-            model_parameters=app_generate_entity.model_conf.parameters,
-            stop=app_generate_entity.model_conf.stop,
-            stream=True,
-        )
-
-        # Consume generator and collect result
-        result: AgentResult | None = None
-        try:
-            while True:
-                try:
-                    output = next(generator)
-                except StopIteration as e:
-                    # Generator finished, get the return value
-                    result = e.value
-                    break
-
-                if isinstance(output, LLMResultChunk):
-                    # Handle LLM chunk
-                    if current_agent_thought_id and not has_published_thought:
-                        self.queue_manager.publish(
-                            QueueAgentThoughtEvent(agent_thought_id=current_agent_thought_id),
-                            PublishFrom.APPLICATION_MANAGER,
-                        )
-                        has_published_thought = True
-
-                    yield output
-
-                elif isinstance(output, AgentLog):
-                    # Handle Agent Log using log_type for type-safe dispatch
-                    if output.status == AgentLog.LogStatus.START:
-                        if output.log_type == AgentLog.LogType.ROUND:
-                            # Start of a new round
-                            message_file_ids: list[str] = []
-                            current_agent_thought_id = self.create_agent_thought(
-                                message_id=message.id,
-                                message="",
-                                tool_name="",
-                                tool_input="",
-                                messages_ids=message_file_ids,
-                            )
-                            has_published_thought = False
-
-                        elif output.log_type == AgentLog.LogType.TOOL_CALL:
-                            if current_agent_thought_id is None:
-                                continue
-
-                            # Tool call start - extract data from structured fields
-                            current_tool_name = output.data.get("tool_name", "")
-                            tool_input = output.data.get("tool_args", {})
-
-                            self.save_agent_thought(
-                                agent_thought_id=current_agent_thought_id,
-                                tool_name=current_tool_name,
-                                tool_input=tool_input,
-                                thought=None,
-                                observation=None,
-                                tool_invoke_meta=None,
-                                answer=None,
-                                messages_ids=[],
-                            )
-                            self.queue_manager.publish(
-                                QueueAgentThoughtEvent(agent_thought_id=current_agent_thought_id),
-                                PublishFrom.APPLICATION_MANAGER,
-                            )
-
-                    elif output.status == AgentLog.LogStatus.SUCCESS:
-                        if output.log_type == AgentLog.LogType.THOUGHT:
-                            if current_agent_thought_id is None:
-                                continue
-
-                            thought_text = output.data.get("thought")
-                            self.save_agent_thought(
-                                agent_thought_id=current_agent_thought_id,
-                                tool_name=None,
-                                tool_input=None,
-                                thought=thought_text,
-                                observation=None,
-                                tool_invoke_meta=None,
-                                answer=None,
-                                messages_ids=[],
-                            )
-                            self.queue_manager.publish(
-                                QueueAgentThoughtEvent(agent_thought_id=current_agent_thought_id),
-                                PublishFrom.APPLICATION_MANAGER,
-                            )
-
-                        elif output.log_type == AgentLog.LogType.TOOL_CALL:
-                            if current_agent_thought_id is None:
-                                continue
-
-                            # Tool call finished
-                            tool_output = output.data.get("output")
-                            # Get meta from strategy output (now properly populated)
-                            tool_meta = output.data.get("meta")
-
-                            # Wrap tool_meta with tool_name as key (required by agent_service)
-                            if tool_meta and current_tool_name:
-                                tool_meta = {current_tool_name: tool_meta}
-
-                            self.save_agent_thought(
-                                agent_thought_id=current_agent_thought_id,
-                                tool_name=None,
-                                tool_input=None,
-                                thought=None,
-                                observation=tool_output,
-                                tool_invoke_meta=tool_meta,
-                                answer=None,
-                                messages_ids=self._current_message_file_ids,
-                            )
-                            # Clear message file ids after saving
-                            self._current_message_file_ids = []
-                            current_tool_name = None
-
-                            self.queue_manager.publish(
-                                QueueAgentThoughtEvent(agent_thought_id=current_agent_thought_id),
-                                PublishFrom.APPLICATION_MANAGER,
-                            )
-
-                        elif output.log_type == AgentLog.LogType.ROUND:
-                            if current_agent_thought_id is None:
-                                continue
-
-                            # Round finished - save LLM usage and answer
-                            llm_usage = output.metadata.get(AgentLog.LogMetadata.LLM_USAGE)
-                            llm_result = output.data.get("llm_result")
-                            final_answer = output.data.get("final_answer")
-
-                            self.save_agent_thought(
-                                agent_thought_id=current_agent_thought_id,
-                                tool_name=None,
-                                tool_input=None,
-                                thought=llm_result,
-                                observation=None,
-                                tool_invoke_meta=None,
-                                answer=final_answer,
-                                messages_ids=[],
-                                llm_usage=llm_usage,
-                            )
-                            self.queue_manager.publish(
-                                QueueAgentThoughtEvent(agent_thought_id=current_agent_thought_id),
-                                PublishFrom.APPLICATION_MANAGER,
-                            )
-
-        except Exception:
-            # Re-raise any other exceptions
-            raise
-
-        # Process final result
-        if isinstance(result, AgentResult):
-            final_answer = result.text
-            usage = result.usage or LLMUsage.empty_usage()
-
-            # Publish end event
-            self.queue_manager.publish(
-                QueueMessageEndEvent(
-                    llm_result=LLMResult(
-                        model=self.model_instance.model,
-                        prompt_messages=prompt_messages,
-                        message=AssistantPromptMessage(content=final_answer),
-                        usage=usage,
-                        system_fingerprint="",
-                    )
-                ),
-                PublishFrom.APPLICATION_MANAGER,
-            )
-
-    def _init_system_message(self, prompt_template: str, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
-        """
-        Initialize system message
-        """
-        if not prompt_template:
-            return prompt_messages or []
-
-        prompt_messages = prompt_messages or []
-
-        if prompt_messages and isinstance(prompt_messages[0], SystemPromptMessage):
-            prompt_messages[0] = SystemPromptMessage(content=prompt_template)
-            return prompt_messages
-
-        if not prompt_messages:
-            return [SystemPromptMessage(content=prompt_template)]
-
-        prompt_messages.insert(0, SystemPromptMessage(content=prompt_template))
-        return prompt_messages
-
-    def _organize_user_query(self, query: str, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
-        """
-        Organize user query
-        """
-        if self.files:
-            # get image detail config
-            image_detail_config = (
-                self.application_generate_entity.file_upload_config.image_config.detail
-                if (
-                    self.application_generate_entity.file_upload_config
-                    and self.application_generate_entity.file_upload_config.image_config
-                )
-                else None
-            )
-            image_detail_config = image_detail_config or ImagePromptMessageContent.DETAIL.LOW
-
-            prompt_message_contents: list[PromptMessageContentUnionTypes] = []
-            for file in self.files:
-                prompt_message_contents.append(
-                    file_manager.to_prompt_message_content(
-                        file,
-                        image_detail_config=image_detail_config,
-                    )
-                )
-            prompt_message_contents.append(TextPromptMessageContent(data=query))
-
-            prompt_messages.append(UserPromptMessage(content=prompt_message_contents))
-        else:
-            prompt_messages.append(UserPromptMessage(content=query))
-
-        return prompt_messages
-
-    def _clear_user_prompt_image_messages(self, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
-        """
-        As for now, gpt supports both fc and vision at the first iteration.
-        We need to remove the image messages from the prompt messages at the first iteration.
-        """
-        prompt_messages = deepcopy(prompt_messages)
-
-        for prompt_message in prompt_messages:
-            if isinstance(prompt_message, UserPromptMessage):
-                if isinstance(prompt_message.content, list):
-                    prompt_message.content = "\n".join(
-                        [
-                            content.data
-                            if content.type == PromptMessageContentType.TEXT
-                            else "[image]"
-                            if content.type == PromptMessageContentType.IMAGE
-                            else "[file]"
-                            for content in prompt_message.content
-                        ]
-                    )
-
-        return prompt_messages
-
-    def _organize_prompt_messages(self):
-        # For ReAct strategy, use the agent prompt template
-        if self.config.strategy == AgentEntity.Strategy.CHAIN_OF_THOUGHT and self.config.prompt:
-            prompt_template = self.config.prompt.first_prompt
-        else:
-            prompt_template = self.app_config.prompt_template.simple_prompt_template or ""
-
-        self.history_prompt_messages = self._init_system_message(prompt_template, self.history_prompt_messages)
-        query_prompt_messages = self._organize_user_query(self.query or "", [])
-
-        self.history_prompt_messages = AgentHistoryPromptTransform(
-            model_config=self.model_config,
-            prompt_messages=[*query_prompt_messages, *self._current_thoughts],
-            history_messages=self.history_prompt_messages,
-            memory=self.memory,
-        ).get_prompt()
-
-        prompt_messages = [*self.history_prompt_messages, *query_prompt_messages, *self._current_thoughts]
-        if len(self._current_thoughts) != 0:
-            # clear messages after the first iteration
-            prompt_messages = self._clear_user_prompt_image_messages(prompt_messages)
-        return prompt_messages
@@ -1,11 +1,12 @@
 import json
 import logging
 import uuid
+from decimal import Decimal
 from typing import Union, cast

 from sqlalchemy import select

-from core.agent.entities import AgentEntity, AgentToolEntity, ExecutionContext
+from core.agent.entities import AgentEntity, AgentToolEntity
 from core.app.app_config.features.file_upload.manager import FileUploadConfigManager
 from core.app.apps.agent_chat.app_config_manager import AgentChatAppConfig
 from core.app.apps.base_app_queue_manager import AppQueueManager
@@ -41,6 +42,7 @@ from core.tools.tool_manager import ToolManager
 from core.tools.utils.dataset_retriever_tool import DatasetRetrieverTool
 from extensions.ext_database import db
 from factories import file_factory
+from models.enums import CreatorUserRole
 from models.model import Conversation, Message, MessageAgentThought, MessageFile

 logger = logging.getLogger(__name__)
@@ -114,20 +116,9 @@ class BaseAgentRunner(AppRunner):
        features = model_schema.features if model_schema and model_schema.features else []
        self.stream_tool_call = ModelFeature.STREAM_TOOL_CALL in features
        self.files = application_generate_entity.files if ModelFeature.VISION in features else []
-        self.model_features = features
        self.query: str | None = ""
        self._current_thoughts: list[PromptMessage] = []

-    def build_execution_context(self) -> ExecutionContext:
-        """Build execution context."""
-        return ExecutionContext(
-            user_id=self.user_id,
-            app_id=self.app_config.app_id,
-            conversation_id=self.conversation.id,
-            message_id=self.message.id,
-            tenant_id=self.tenant_id,
-        )
-
    def _repack_app_generate_entity(
        self, app_generate_entity: AgentChatAppGenerateEntity
    ) -> AgentChatAppGenerateEntity:
@@ -300,6 +291,7 @@ class BaseAgentRunner(AppRunner):
        thought = MessageAgentThought(
            message_id=message_id,
            message_chain_id=None,
+            tool_process_data=None,
            thought="",
            tool=tool_name,
            tool_labels_str="{}",
@@ -307,20 +299,20 @@ class BaseAgentRunner(AppRunner):
            tool_input=tool_input,
            message=message,
            message_token=0,
-            message_unit_price=0,
-            message_price_unit=0,
+            message_unit_price=Decimal(0),
+            message_price_unit=Decimal("0.001"),
            message_files=json.dumps(messages_ids) if messages_ids else "",
            answer="",
            observation="",
            answer_token=0,
-            answer_unit_price=0,
-            answer_price_unit=0,
+            answer_unit_price=Decimal(0),
+            answer_price_unit=Decimal("0.001"),
            tokens=0,
-            total_price=0,
+            total_price=Decimal(0),
            position=self.agent_thought_count + 1,
            currency="USD",
            latency=0,
-            created_by_role="account",
+            created_by_role=CreatorUserRole.ACCOUNT,
            created_by=self.user_id,
        )

@@ -353,7 +345,8 @@ class BaseAgentRunner(AppRunner):
            raise ValueError("agent thought not found")

        if thought:
-            agent_thought.thought += thought
+            existing_thought = agent_thought.thought or ""
+            agent_thought.thought = f"{existing_thought}{thought}"

        if tool_name:
            agent_thought.tool = tool_name
@@ -451,21 +444,30 @@ class BaseAgentRunner(AppRunner):
            agent_thoughts: list[MessageAgentThought] = message.agent_thoughts
            if agent_thoughts:
                for agent_thought in agent_thoughts:
-                    tools = agent_thought.tool
-                    if tools:
-                        tools = tools.split(";")
+                    tool_names_raw = agent_thought.tool
+                    if tool_names_raw:
+                        tool_names = tool_names_raw.split(";")
                        tool_calls: list[AssistantPromptMessage.ToolCall] = []
                        tool_call_response: list[ToolPromptMessage] = []
-                        try:
-                            tool_inputs = json.loads(agent_thought.tool_input)
-                        except Exception:
-                            tool_inputs = {tool: {} for tool in tools}
-                        try:
-                            tool_responses = json.loads(agent_thought.observation)
-                        except Exception:
-                            tool_responses = dict.fromkeys(tools, agent_thought.observation)
+                        tool_input_payload = agent_thought.tool_input
+                        if tool_input_payload:
+                            try:
+                                tool_inputs = json.loads(tool_input_payload)
+                            except Exception:
+                                tool_inputs = {tool: {} for tool in tool_names}
+                        else:
+                            tool_inputs = {tool: {} for tool in tool_names}

-                        for tool in tools:
+                        observation_payload = agent_thought.observation
+                        if observation_payload:
+                            try:
+                                tool_responses = json.loads(observation_payload)
+                            except Exception:
+                                tool_responses = dict.fromkeys(tool_names, observation_payload)
+                        else:
+                            tool_responses = dict.fromkeys(tool_names, observation_payload)
+
+                        for tool in tool_names:
                            # generate a uuid for tool call
                            tool_call_id = str(uuid.uuid4())
                            tool_calls.append(
@@ -495,7 +497,7 @@ class BaseAgentRunner(AppRunner):
                                *tool_call_response,
                            ]
                        )
-                    if not tools:
+                    if not tool_names_raw:
                        result.append(AssistantPromptMessage(content=agent_thought.thought))
            else:
                if message.answer:
@@ -0,0 +1,437 @@
+import json
+import logging
+from abc import ABC, abstractmethod
+from collections.abc import Generator, Mapping, Sequence
+from typing import Any
+
+from core.agent.base_agent_runner import BaseAgentRunner
+from core.agent.entities import AgentScratchpadUnit
+from core.agent.output_parser.cot_output_parser import CotAgentOutputParser
+from core.app.apps.base_app_queue_manager import PublishFrom
+from core.app.entities.queue_entities import QueueAgentThoughtEvent, QueueMessageEndEvent, QueueMessageFileEvent
+from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
+from core.model_runtime.entities.message_entities import (
+    AssistantPromptMessage,
+    PromptMessage,
+    PromptMessageTool,
+    ToolPromptMessage,
+    UserPromptMessage,
+)
+from core.ops.ops_trace_manager import TraceQueueManager
+from core.prompt.agent_history_prompt_transform import AgentHistoryPromptTransform
+from core.tools.__base.tool import Tool
+from core.tools.entities.tool_entities import ToolInvokeMeta
+from core.tools.tool_engine import ToolEngine
+from core.workflow.nodes.agent.exc import AgentMaxIterationError
+from models.model import Message
+
+logger = logging.getLogger(__name__)
+
+
+class CotAgentRunner(BaseAgentRunner, ABC):
+    _is_first_iteration = True
+    _ignore_observation_providers = ["wenxin"]
+    _historic_prompt_messages: list[PromptMessage]
+    _agent_scratchpad: list[AgentScratchpadUnit]
+    _instruction: str
+    _query: str
+    _prompt_messages_tools: Sequence[PromptMessageTool]
+
+    def run(
+        self,
+        message: Message,
+        query: str,
+        inputs: Mapping[str, str],
+    ) -> Generator:
+        """
+        Run Cot agent application
+        """
+
+        app_generate_entity = self.application_generate_entity
+        self._repack_app_generate_entity(app_generate_entity)
+        self._init_react_state(query)
+
+        trace_manager = app_generate_entity.trace_manager
+
+        # check model mode
+        if "Observation" not in app_generate_entity.model_conf.stop:
+            if app_generate_entity.model_conf.provider not in self._ignore_observation_providers:
+                app_generate_entity.model_conf.stop.append("Observation")
+
+        app_config = self.app_config
+        assert app_config.agent
+
+        # init instruction
+        inputs = inputs or {}
+        instruction = app_config.prompt_template.simple_prompt_template or ""
+        self._instruction = self._fill_in_inputs_from_external_data_tools(instruction, inputs)
+
+        iteration_step = 1
+        max_iteration_steps = min(app_config.agent.max_iteration, 99) + 1
+
+        # convert tools into ModelRuntime Tool format
+        tool_instances, prompt_messages_tools = self._init_prompt_tools()
+        self._prompt_messages_tools = prompt_messages_tools
+
+        function_call_state = True
+        llm_usage: dict[str, LLMUsage | None] = {"usage": None}
+        final_answer = ""
+        prompt_messages: list = []  # Initialize prompt_messages
+        agent_thought_id = ""  # Initialize agent_thought_id
+
+        def increase_usage(final_llm_usage_dict: dict[str, LLMUsage | None], usage: LLMUsage):
+            if not final_llm_usage_dict["usage"]:
+                final_llm_usage_dict["usage"] = usage
+            else:
+                llm_usage = final_llm_usage_dict["usage"]
+                llm_usage.prompt_tokens += usage.prompt_tokens
+                llm_usage.completion_tokens += usage.completion_tokens
+                llm_usage.total_tokens += usage.total_tokens
+                llm_usage.prompt_price += usage.prompt_price
+                llm_usage.completion_price += usage.completion_price
+                llm_usage.total_price += usage.total_price
+
+        model_instance = self.model_instance
+
+        while function_call_state and iteration_step <= max_iteration_steps:
+            # continue to run until there is not any tool call
+            function_call_state = False
+
+            if iteration_step == max_iteration_steps:
+                # the last iteration, remove all tools
+                self._prompt_messages_tools = []
+
+            message_file_ids: list[str] = []
+
+            agent_thought_id = self.create_agent_thought(
+                message_id=message.id, message="", tool_name="", tool_input="", messages_ids=message_file_ids
+            )
+
+            if iteration_step > 1:
+                self.queue_manager.publish(
+                    QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                )
+
+            # recalc llm max tokens
+            prompt_messages = self._organize_prompt_messages()
+            self.recalc_llm_max_tokens(self.model_config, prompt_messages)
+            # invoke model
+            chunks = model_instance.invoke_llm(
+                prompt_messages=prompt_messages,
+                model_parameters=app_generate_entity.model_conf.parameters,
+                tools=[],
+                stop=app_generate_entity.model_conf.stop,
+                stream=True,
+                user=self.user_id,
+                callbacks=[],
+            )
+
+            usage_dict: dict[str, LLMUsage | None] = {}
+            react_chunks = CotAgentOutputParser.handle_react_stream_output(chunks, usage_dict)
+            scratchpad = AgentScratchpadUnit(
+                agent_response="",
+                thought="",
+                action_str="",
+                observation="",
+                action=None,
+            )
+
+            # publish agent thought if it's first iteration
+            if iteration_step == 1:
+                self.queue_manager.publish(
+                    QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                )
+
+            for chunk in react_chunks:
+                if isinstance(chunk, AgentScratchpadUnit.Action):
+                    action = chunk
+                    # detect action
+                    assert scratchpad.agent_response is not None
+                    scratchpad.agent_response += json.dumps(chunk.model_dump())
+                    scratchpad.action_str = json.dumps(chunk.model_dump())
+                    scratchpad.action = action
+                else:
+                    assert scratchpad.agent_response is not None
+                    scratchpad.agent_response += chunk
+                    assert scratchpad.thought is not None
+                    scratchpad.thought += chunk
+                    yield LLMResultChunk(
+                        model=self.model_config.model,
+                        prompt_messages=prompt_messages,
+                        system_fingerprint="",
+                        delta=LLMResultChunkDelta(index=0, message=AssistantPromptMessage(content=chunk), usage=None),
+                    )
+
+            assert scratchpad.thought is not None
+            scratchpad.thought = scratchpad.thought.strip() or "I am thinking about how to help you"
+            self._agent_scratchpad.append(scratchpad)
+
+            # Check if max iteration is reached and model still wants to call tools
+            if iteration_step == max_iteration_steps and scratchpad.action:
+                if scratchpad.action.action_name.lower() != "final answer":
+                    raise AgentMaxIterationError(app_config.agent.max_iteration)
+
+            # get llm usage
+            if "usage" in usage_dict:
+                if usage_dict["usage"] is not None:
+                    increase_usage(llm_usage, usage_dict["usage"])
+            else:
+                usage_dict["usage"] = LLMUsage.empty_usage()
+
+            self.save_agent_thought(
+                agent_thought_id=agent_thought_id,
+                tool_name=(scratchpad.action.action_name if scratchpad.action and not scratchpad.is_final() else ""),
+                tool_input={scratchpad.action.action_name: scratchpad.action.action_input} if scratchpad.action else {},
+                tool_invoke_meta={},
+                thought=scratchpad.thought or "",
+                observation="",
+                answer=scratchpad.agent_response or "",
+                messages_ids=[],
+                llm_usage=usage_dict["usage"],
+            )
+
+            if not scratchpad.is_final():
+                self.queue_manager.publish(
+                    QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                )
+
+            if not scratchpad.action:
+                # failed to extract action, return final answer directly
+                final_answer = ""
+            else:
+                if scratchpad.action.action_name.lower() == "final answer":
+                    # action is final answer, return final answer directly
+                    try:
+                        if isinstance(scratchpad.action.action_input, dict):
+                            final_answer = json.dumps(scratchpad.action.action_input, ensure_ascii=False)
+                        elif isinstance(scratchpad.action.action_input, str):
+                            final_answer = scratchpad.action.action_input
+                        else:
+                            final_answer = f"{scratchpad.action.action_input}"
+                    except TypeError:
+                        final_answer = f"{scratchpad.action.action_input}"
+                else:
+                    function_call_state = True
+                    # action is tool call, invoke tool
+                    tool_invoke_response, tool_invoke_meta = self._handle_invoke_action(
+                        action=scratchpad.action,
+                        tool_instances=tool_instances,
+                        message_file_ids=message_file_ids,
+                        trace_manager=trace_manager,
+                    )
+                    scratchpad.observation = tool_invoke_response
+                    scratchpad.agent_response = tool_invoke_response
+
+                    self.save_agent_thought(
+                        agent_thought_id=agent_thought_id,
+                        tool_name=scratchpad.action.action_name,
+                        tool_input={scratchpad.action.action_name: scratchpad.action.action_input},
+                        thought=scratchpad.thought or "",
+                        observation={scratchpad.action.action_name: tool_invoke_response},
+                        tool_invoke_meta={scratchpad.action.action_name: tool_invoke_meta.to_dict()},
+                        answer=scratchpad.agent_response,
+                        messages_ids=message_file_ids,
+                        llm_usage=usage_dict["usage"],
+                    )
+
+                    self.queue_manager.publish(
+                        QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                    )
+
+                # update prompt tool message
+                for prompt_tool in self._prompt_messages_tools:
+                    self.update_prompt_message_tool(tool_instances[prompt_tool.name], prompt_tool)
+
+            iteration_step += 1
+
+        yield LLMResultChunk(
+            model=model_instance.model,
+            prompt_messages=prompt_messages,
+            delta=LLMResultChunkDelta(
+                index=0, message=AssistantPromptMessage(content=final_answer), usage=llm_usage["usage"]
+            ),
+            system_fingerprint="",
+        )
+
+        # save agent thought
+        self.save_agent_thought(
+            agent_thought_id=agent_thought_id,
+            tool_name="",
+            tool_input={},
+            tool_invoke_meta={},
+            thought=final_answer,
+            observation={},
+            answer=final_answer,
+            messages_ids=[],
+        )
+        # publish end event
+        self.queue_manager.publish(
+            QueueMessageEndEvent(
+                llm_result=LLMResult(
+                    model=model_instance.model,
+                    prompt_messages=prompt_messages,
+                    message=AssistantPromptMessage(content=final_answer),
+                    usage=llm_usage["usage"] or LLMUsage.empty_usage(),
+                    system_fingerprint="",
+                )
+            ),
+            PublishFrom.APPLICATION_MANAGER,
+        )
+
+    def _handle_invoke_action(
+        self,
+        action: AgentScratchpadUnit.Action,
+        tool_instances: Mapping[str, Tool],
+        message_file_ids: list[str],
+        trace_manager: TraceQueueManager | None = None,
+    ) -> tuple[str, ToolInvokeMeta]:
+        """
+        handle invoke action
+        :param action: action
+        :param tool_instances: tool instances
+        :param message_file_ids: message file ids
+        :param trace_manager: trace manager
+        :return: observation, meta
+        """
+        # action is tool call, invoke tool
+        tool_call_name = action.action_name
+        tool_call_args = action.action_input
+        tool_instance = tool_instances.get(tool_call_name)
+
+        if not tool_instance:
+            answer = f"there is not a tool named {tool_call_name}"
+            return answer, ToolInvokeMeta.error_instance(answer)
+
+        if isinstance(tool_call_args, str):
+            try:
+                tool_call_args = json.loads(tool_call_args)
+            except json.JSONDecodeError:
+                pass
+
+        # invoke tool
+        tool_invoke_response, message_files, tool_invoke_meta = ToolEngine.agent_invoke(
+            tool=tool_instance,
+            tool_parameters=tool_call_args,
+            user_id=self.user_id,
+            tenant_id=self.tenant_id,
+            message=self.message,
+            invoke_from=self.application_generate_entity.invoke_from,
+            agent_tool_callback=self.agent_callback,
+            trace_manager=trace_manager,
+        )
+
+        # publish files
+        for message_file_id in message_files:
+            # publish message file
+            self.queue_manager.publish(
+                QueueMessageFileEvent(message_file_id=message_file_id), PublishFrom.APPLICATION_MANAGER
+            )
+            # add message file ids
+            message_file_ids.append(message_file_id)
+
+        return tool_invoke_response, tool_invoke_meta
+
+    def _convert_dict_to_action(self, action: dict) -> AgentScratchpadUnit.Action:
+        """
+        convert dict to action
+        """
+        return AgentScratchpadUnit.Action(action_name=action["action"], action_input=action["action_input"])
+
+    def _fill_in_inputs_from_external_data_tools(self, instruction: str, inputs: Mapping[str, Any]) -> str:
+        """
+        fill in inputs from external data tools
+        """
+        for key, value in inputs.items():
+            try:
+                instruction = instruction.replace(f"{{{{{key}}}}}", str(value))
+            except Exception:
+                continue
+
+        return instruction
+
+    def _init_react_state(self, query):
+        """
+        init agent scratchpad
+        """
+        self._query = query
+        self._agent_scratchpad = []
+        self._historic_prompt_messages = self._organize_historic_prompt_messages()
+
+    @abstractmethod
+    def _organize_prompt_messages(self) -> list[PromptMessage]:
+        """
+        organize prompt messages
+        """
+
+    def _format_assistant_message(self, agent_scratchpad: list[AgentScratchpadUnit]) -> str:
+        """
+        format assistant message
+        """
+        message = ""
+        for scratchpad in agent_scratchpad:
+            if scratchpad.is_final():
+                message += f"Final Answer: {scratchpad.agent_response}"
+            else:
+                message += f"Thought: {scratchpad.thought}\n\n"
+                if scratchpad.action_str:
+                    message += f"Action: {scratchpad.action_str}\n\n"
+                if scratchpad.observation:
+                    message += f"Observation: {scratchpad.observation}\n\n"
+
+        return message
+
+    def _organize_historic_prompt_messages(
+        self, current_session_messages: list[PromptMessage] | None = None
+    ) -> list[PromptMessage]:
+        """
+        organize historic prompt messages
+        """
+        result: list[PromptMessage] = []
+        scratchpads: list[AgentScratchpadUnit] = []
+        current_scratchpad: AgentScratchpadUnit | None = None
+
+        for message in self.history_prompt_messages:
+            if isinstance(message, AssistantPromptMessage):
+                if not current_scratchpad:
+                    assert isinstance(message.content, str)
+                    current_scratchpad = AgentScratchpadUnit(
+                        agent_response=message.content,
+                        thought=message.content or "I am thinking about how to help you",
+                        action_str="",
+                        action=None,
+                        observation=None,
+                    )
+                    scratchpads.append(current_scratchpad)
+                if message.tool_calls:
+                    try:
+                        current_scratchpad.action = AgentScratchpadUnit.Action(
+                            action_name=message.tool_calls[0].function.name,
+                            action_input=json.loads(message.tool_calls[0].function.arguments),
+                        )
+                        current_scratchpad.action_str = json.dumps(current_scratchpad.action.to_dict())
+                    except Exception:
+                        logger.exception("Failed to parse tool call from assistant message")
+            elif isinstance(message, ToolPromptMessage):
+                if current_scratchpad:
+                    assert isinstance(message.content, str)
+                    current_scratchpad.observation = message.content
+                else:
+                    raise NotImplementedError("expected str type")
+            elif isinstance(message, UserPromptMessage):
+                if scratchpads:
+                    result.append(AssistantPromptMessage(content=self._format_assistant_message(scratchpads)))
+                    scratchpads = []
+                    current_scratchpad = None
+
+                result.append(message)
+
+        if scratchpads:
+            result.append(AssistantPromptMessage(content=self._format_assistant_message(scratchpads)))
+
+        historic_prompts = AgentHistoryPromptTransform(
+            model_config=self.model_config,
+            prompt_messages=current_session_messages or [],
+            history_messages=result,
+            memory=self.memory,
+        ).get_prompt()
+        return historic_prompts
@@ -0,0 +1,118 @@
+import json
+
+from core.agent.cot_agent_runner import CotAgentRunner
+from core.file import file_manager
+from core.model_runtime.entities import (
+    AssistantPromptMessage,
+    PromptMessage,
+    SystemPromptMessage,
+    TextPromptMessageContent,
+    UserPromptMessage,
+)
+from core.model_runtime.entities.message_entities import ImagePromptMessageContent, PromptMessageContentUnionTypes
+from core.model_runtime.utils.encoders import jsonable_encoder
+
+
+class CotChatAgentRunner(CotAgentRunner):
+    def _organize_system_prompt(self) -> SystemPromptMessage:
+        """
+        Organize system prompt
+        """
+        assert self.app_config.agent
+        assert self.app_config.agent.prompt
+
+        prompt_entity = self.app_config.agent.prompt
+        if not prompt_entity:
+            raise ValueError("Agent prompt configuration is not set")
+        first_prompt = prompt_entity.first_prompt
+
+        system_prompt = (
+            first_prompt.replace("{{instruction}}", self._instruction)
+            .replace("{{tools}}", json.dumps(jsonable_encoder(self._prompt_messages_tools)))
+            .replace("{{tool_names}}", ", ".join([tool.name for tool in self._prompt_messages_tools]))
+        )
+
+        return SystemPromptMessage(content=system_prompt)
+
+    def _organize_user_query(self, query, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
+        """
+        Organize user query
+        """
+        if self.files:
+            # get image detail config
+            image_detail_config = (
+                self.application_generate_entity.file_upload_config.image_config.detail
+                if (
+                    self.application_generate_entity.file_upload_config
+                    and self.application_generate_entity.file_upload_config.image_config
+                )
+                else None
+            )
+            image_detail_config = image_detail_config or ImagePromptMessageContent.DETAIL.LOW
+
+            prompt_message_contents: list[PromptMessageContentUnionTypes] = []
+            for file in self.files:
+                prompt_message_contents.append(
+                    file_manager.to_prompt_message_content(
+                        file,
+                        image_detail_config=image_detail_config,
+                    )
+                )
+            prompt_message_contents.append(TextPromptMessageContent(data=query))
+
+            prompt_messages.append(UserPromptMessage(content=prompt_message_contents))
+        else:
+            prompt_messages.append(UserPromptMessage(content=query))
+
+        return prompt_messages
+
+    def _organize_prompt_messages(self) -> list[PromptMessage]:
+        """
+        Organize
+        """
+        # organize system prompt
+        system_message = self._organize_system_prompt()
+
+        # organize current assistant messages
+        agent_scratchpad = self._agent_scratchpad
+        if not agent_scratchpad:
+            assistant_messages = []
+        else:
+            assistant_message = AssistantPromptMessage(content="")
+            assistant_message.content = ""  # FIXME: type check tell mypy that assistant_message.content is str
+            for unit in agent_scratchpad:
+                if unit.is_final():
+                    assert isinstance(assistant_message.content, str)
+                    assistant_message.content += f"Final Answer: {unit.agent_response}"
+                else:
+                    assert isinstance(assistant_message.content, str)
+                    assistant_message.content += f"Thought: {unit.thought}\n\n"
+                    if unit.action_str:
+                        assistant_message.content += f"Action: {unit.action_str}\n\n"
+                    if unit.observation:
+                        assistant_message.content += f"Observation: {unit.observation}\n\n"
+
+            assistant_messages = [assistant_message]
+
+        # query messages
+        query_messages = self._organize_user_query(self._query, [])
+
+        if assistant_messages:
+            # organize historic prompt messages
+            historic_messages = self._organize_historic_prompt_messages(
+                [system_message, *query_messages, *assistant_messages, UserPromptMessage(content="continue")]
+            )
+            messages = [
+                system_message,
+                *historic_messages,
+                *query_messages,
+                *assistant_messages,
+                UserPromptMessage(content="continue"),
+            ]
+        else:
+            # organize historic prompt messages
+            historic_messages = self._organize_historic_prompt_messages([system_message, *query_messages])
+            messages = [system_message, *historic_messages, *query_messages]
+
+        # join all messages
+        return messages
@@ -0,0 +1,87 @@
+import json
+
+from core.agent.cot_agent_runner import CotAgentRunner
+from core.model_runtime.entities.message_entities import (
+    AssistantPromptMessage,
+    PromptMessage,
+    TextPromptMessageContent,
+    UserPromptMessage,
+)
+from core.model_runtime.utils.encoders import jsonable_encoder
+
+
+class CotCompletionAgentRunner(CotAgentRunner):
+    def _organize_instruction_prompt(self) -> str:
+        """
+        Organize instruction prompt
+        """
+        if self.app_config.agent is None:
+            raise ValueError("Agent configuration is not set")
+        prompt_entity = self.app_config.agent.prompt
+        if prompt_entity is None:
+            raise ValueError("prompt entity is not set")
+        first_prompt = prompt_entity.first_prompt
+
+        system_prompt = (
+            first_prompt.replace("{{instruction}}", self._instruction)
+            .replace("{{tools}}", json.dumps(jsonable_encoder(self._prompt_messages_tools)))
+            .replace("{{tool_names}}", ", ".join([tool.name for tool in self._prompt_messages_tools]))
+        )
+
+        return system_prompt
+
+    def _organize_historic_prompt(self, current_session_messages: list[PromptMessage] | None = None) -> str:
+        """
+        Organize historic prompt
+        """
+        historic_prompt_messages = self._organize_historic_prompt_messages(current_session_messages)
+        historic_prompt = ""
+
+        for message in historic_prompt_messages:
+            if isinstance(message, UserPromptMessage):
+                historic_prompt += f"Question: {message.content}\n\n"
+            elif isinstance(message, AssistantPromptMessage):
+                if isinstance(message.content, str):
+                    historic_prompt += message.content + "\n\n"
+                elif isinstance(message.content, list):
+                    for content in message.content:
+                        if not isinstance(content, TextPromptMessageContent):
+                            continue
+                        historic_prompt += content.data
+
+        return historic_prompt
+
+    def _organize_prompt_messages(self) -> list[PromptMessage]:
+        """
+        Organize prompt messages
+        """
+        # organize system prompt
+        system_prompt = self._organize_instruction_prompt()
+
+        # organize historic prompt messages
+        historic_prompt = self._organize_historic_prompt()
+
+        # organize current assistant messages
+        agent_scratchpad = self._agent_scratchpad
+        assistant_prompt = ""
+        for unit in agent_scratchpad or []:
+            if unit.is_final():
+                assistant_prompt += f"Final Answer: {unit.agent_response}"
+            else:
+                assistant_prompt += f"Thought: {unit.thought}\n\n"
+                if unit.action_str:
+                    assistant_prompt += f"Action: {unit.action_str}\n\n"
+                if unit.observation:
+                    assistant_prompt += f"Observation: {unit.observation}\n\n"
+
+        # query messages
+        query_prompt = f"Question: {self._query}"
+
+        # join all messages
+        prompt = (
+            system_prompt.replace("{{historic_messages}}", historic_prompt)
+            .replace("{{agent_scratchpad}}", assistant_prompt)
+            .replace("{{query}}", query_prompt)
+        )
+
+        return [UserPromptMessage(content=prompt)]
@@ -1,5 +1,3 @@
-import uuid
-from collections.abc import Mapping
 from enum import StrEnum
 from typing import Any, Union

@@ -94,96 +92,3 @@ class AgentInvokeMessage(ToolInvokeMessage):
    """

    pass
-
-
-class ExecutionContext(BaseModel):
-    """Execution context containing trace and audit information.
-
-    This context carries all the IDs and metadata that are not part of
-    the core business logic but needed for tracing, auditing, and
-    correlation purposes.
-    """
-
-    user_id: str | None = None
-    app_id: str | None = None
-    conversation_id: str | None = None
-    message_id: str | None = None
-    tenant_id: str | None = None
-
-    @classmethod
-    def create_minimal(cls, user_id: str | None = None) -> "ExecutionContext":
-        """Create a minimal context with only essential fields."""
-        return cls(user_id=user_id)
-
-    def to_dict(self) -> dict[str, Any]:
-        """Convert to dictionary for passing to legacy code."""
-        return {
-            "user_id": self.user_id,
-            "app_id": self.app_id,
-            "conversation_id": self.conversation_id,
-            "message_id": self.message_id,
-            "tenant_id": self.tenant_id,
-        }
-
-    def with_updates(self, **kwargs) -> "ExecutionContext":
-        """Create a new context with updated fields."""
-        data = self.to_dict()
-        data.update(kwargs)
-
-        return ExecutionContext(
-            user_id=data.get("user_id"),
-            app_id=data.get("app_id"),
-            conversation_id=data.get("conversation_id"),
-            message_id=data.get("message_id"),
-            tenant_id=data.get("tenant_id"),
-        )
-
-
-class AgentLog(BaseModel):
-    """
-    Agent Log.
-    """
-
-    class LogType(StrEnum):
-        """Type of agent log entry."""
-
-        ROUND = "round"  # A complete iteration round
-        THOUGHT = "thought"  # LLM thinking/reasoning
-        TOOL_CALL = "tool_call"  # Tool invocation
-
-    class LogMetadata(StrEnum):
-        STARTED_AT = "started_at"
-        FINISHED_AT = "finished_at"
-        ELAPSED_TIME = "elapsed_time"
-        TOTAL_PRICE = "total_price"
-        TOTAL_TOKENS = "total_tokens"
-        PROVIDER = "provider"
-        CURRENCY = "currency"
-        LLM_USAGE = "llm_usage"
-        ICON = "icon"
-        ICON_DARK = "icon_dark"
-
-    class LogStatus(StrEnum):
-        START = "start"
-        ERROR = "error"
-        SUCCESS = "success"
-
-    id: str = Field(default_factory=lambda: str(uuid.uuid4()), description="The id of the log")
-    label: str = Field(..., description="The label of the log")
-    log_type: LogType = Field(..., description="The type of the log")
-    parent_id: str | None = Field(default=None, description="Leave empty for root log")
-    error: str | None = Field(default=None, description="The error message")
-    status: LogStatus = Field(..., description="The status of the log")
-    data: Mapping[str, Any] = Field(..., description="Detailed log data")
-    metadata: Mapping[LogMetadata, Any] = Field(default={}, description="The metadata of the log")
-
-
-class AgentResult(BaseModel):
-    """
-    Agent execution result.
-    """
-
-    text: str = Field(default="", description="The generated text")
-    files: list[Any] = Field(default_factory=list, description="Files produced during execution")
-    usage: Any | None = Field(default=None, description="LLM usage statistics")
-    finish_reason: str | None = Field(default=None, description="Reason for completion")
@@ -0,0 +1,468 @@
+import json
+import logging
+from collections.abc import Generator
+from copy import deepcopy
+from typing import Any, Union
+
+from core.agent.base_agent_runner import BaseAgentRunner
+from core.app.apps.base_app_queue_manager import PublishFrom
+from core.app.entities.queue_entities import QueueAgentThoughtEvent, QueueMessageEndEvent, QueueMessageFileEvent
+from core.file import file_manager
+from core.model_runtime.entities import (
+    AssistantPromptMessage,
+    LLMResult,
+    LLMResultChunk,
+    LLMResultChunkDelta,
+    LLMUsage,
+    PromptMessage,
+    PromptMessageContentType,
+    SystemPromptMessage,
+    TextPromptMessageContent,
+    ToolPromptMessage,
+    UserPromptMessage,
+)
+from core.model_runtime.entities.message_entities import ImagePromptMessageContent, PromptMessageContentUnionTypes
+from core.prompt.agent_history_prompt_transform import AgentHistoryPromptTransform
+from core.tools.entities.tool_entities import ToolInvokeMeta
+from core.tools.tool_engine import ToolEngine
+from core.workflow.nodes.agent.exc import AgentMaxIterationError
+from models.model import Message
+
+logger = logging.getLogger(__name__)
+
+
+class FunctionCallAgentRunner(BaseAgentRunner):
+    def run(self, message: Message, query: str, **kwargs: Any) -> Generator[LLMResultChunk, None, None]:
+        """
+        Run FunctionCall agent application
+        """
+        self.query = query
+        app_generate_entity = self.application_generate_entity
+
+        app_config = self.app_config
+        assert app_config is not None, "app_config is required"
+        assert app_config.agent is not None, "app_config.agent is required"
+
+        # convert tools into ModelRuntime Tool format
+        tool_instances, prompt_messages_tools = self._init_prompt_tools()
+
+        assert app_config.agent
+
+        iteration_step = 1
+        max_iteration_steps = min(app_config.agent.max_iteration, 99) + 1
+
+        # continue to run until there is not any tool call
+        function_call_state = True
+        llm_usage: dict[str, LLMUsage | None] = {"usage": None}
+        final_answer = ""
+        prompt_messages: list = []  # Initialize prompt_messages
+
+        # get tracing instance
+        trace_manager = app_generate_entity.trace_manager
+
+        def increase_usage(final_llm_usage_dict: dict[str, LLMUsage | None], usage: LLMUsage):
+            if not final_llm_usage_dict["usage"]:
+                final_llm_usage_dict["usage"] = usage
+            else:
+                llm_usage = final_llm_usage_dict["usage"]
+                llm_usage.prompt_tokens += usage.prompt_tokens
+                llm_usage.completion_tokens += usage.completion_tokens
+                llm_usage.total_tokens += usage.total_tokens
+                llm_usage.prompt_price += usage.prompt_price
+                llm_usage.completion_price += usage.completion_price
+                llm_usage.total_price += usage.total_price
+
+        model_instance = self.model_instance
+
+        while function_call_state and iteration_step <= max_iteration_steps:
+            function_call_state = False
+
+            if iteration_step == max_iteration_steps:
+                # the last iteration, remove all tools
+                prompt_messages_tools = []
+
+            message_file_ids: list[str] = []
+            agent_thought_id = self.create_agent_thought(
+                message_id=message.id, message="", tool_name="", tool_input="", messages_ids=message_file_ids
+            )
+
+            # recalc llm max tokens
+            prompt_messages = self._organize_prompt_messages()
+            self.recalc_llm_max_tokens(self.model_config, prompt_messages)
+            # invoke model
+            chunks: Union[Generator[LLMResultChunk, None, None], LLMResult] = model_instance.invoke_llm(
+                prompt_messages=prompt_messages,
+                model_parameters=app_generate_entity.model_conf.parameters,
+                tools=prompt_messages_tools,
+                stop=app_generate_entity.model_conf.stop,
+                stream=self.stream_tool_call,
+                user=self.user_id,
+                callbacks=[],
+            )
+
+            tool_calls: list[tuple[str, str, dict[str, Any]]] = []
+
+            # save full response
+            response = ""
+
+            # save tool call names and inputs
+            tool_call_names = ""
+            tool_call_inputs = ""
+
+            current_llm_usage = None
+
+            if isinstance(chunks, Generator):
+                is_first_chunk = True
+                for chunk in chunks:
+                    if is_first_chunk:
+                        self.queue_manager.publish(
+                            QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                        )
+                        is_first_chunk = False
+                    # check if there is any tool call
+                    if self.check_tool_calls(chunk):
+                        function_call_state = True
+                        tool_calls.extend(self.extract_tool_calls(chunk) or [])
+                        tool_call_names = ";".join([tool_call[1] for tool_call in tool_calls])
+                        try:
+                            tool_call_inputs = json.dumps(
+                                {tool_call[1]: tool_call[2] for tool_call in tool_calls}, ensure_ascii=False
+                            )
+                        except TypeError:
+                            # fallback: force ASCII to handle non-serializable objects
+                            tool_call_inputs = json.dumps({tool_call[1]: tool_call[2] for tool_call in tool_calls})
+
+                    if chunk.delta.message and chunk.delta.message.content:
+                        if isinstance(chunk.delta.message.content, list):
+                            for content in chunk.delta.message.content:
+                                response += content.data
+                        else:
+                            response += str(chunk.delta.message.content)
+
+                    if chunk.delta.usage:
+                        increase_usage(llm_usage, chunk.delta.usage)
+                        current_llm_usage = chunk.delta.usage
+
+                    yield chunk
+            else:
+                result = chunks
+                # check if there is any tool call
+                if self.check_blocking_tool_calls(result):
+                    function_call_state = True
+                    tool_calls.extend(self.extract_blocking_tool_calls(result) or [])
+                    tool_call_names = ";".join([tool_call[1] for tool_call in tool_calls])
+                    try:
+                        tool_call_inputs = json.dumps(
+                            {tool_call[1]: tool_call[2] for tool_call in tool_calls}, ensure_ascii=False
+                        )
+                    except TypeError:
+                        # fallback: force ASCII to handle non-serializable objects
+                        tool_call_inputs = json.dumps({tool_call[1]: tool_call[2] for tool_call in tool_calls})
+
+                if result.usage:
+                    increase_usage(llm_usage, result.usage)
+                    current_llm_usage = result.usage
+
+                if result.message and result.message.content:
+                    if isinstance(result.message.content, list):
+                        for content in result.message.content:
+                            response += content.data
+                    else:
+                        response += str(result.message.content)
+
+                if not result.message.content:
+                    result.message.content = ""
+
+                self.queue_manager.publish(
+                    QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                )
+
+                yield LLMResultChunk(
+                    model=model_instance.model,
+                    prompt_messages=result.prompt_messages,
+                    system_fingerprint=result.system_fingerprint,
+                    delta=LLMResultChunkDelta(
+                        index=0,
+                        message=result.message,
+                        usage=result.usage,
+                    ),
+                )
+
+            assistant_message = AssistantPromptMessage(content=response, tool_calls=[])
+            if tool_calls:
+                assistant_message.tool_calls = [
+                    AssistantPromptMessage.ToolCall(
+                        id=tool_call[0],
+                        type="function",
+                        function=AssistantPromptMessage.ToolCall.ToolCallFunction(
+                            name=tool_call[1], arguments=json.dumps(tool_call[2], ensure_ascii=False)
+                        ),
+                    )
+                    for tool_call in tool_calls
+                ]
+
+            self._current_thoughts.append(assistant_message)
+
+            # save thought
+            self.save_agent_thought(
+                agent_thought_id=agent_thought_id,
+                tool_name=tool_call_names,
+                tool_input=tool_call_inputs,
+                thought=response,
+                tool_invoke_meta=None,
+                observation=None,
+                answer=response,
+                messages_ids=[],
+                llm_usage=current_llm_usage,
+            )
+            self.queue_manager.publish(
+                QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+            )
+
+            final_answer += response + "\n"
+
+            # Check if max iteration is reached and model still wants to call tools
+            if iteration_step == max_iteration_steps and tool_calls:
+                raise AgentMaxIterationError(app_config.agent.max_iteration)
+
+            # call tools
+            tool_responses = []
+            for tool_call_id, tool_call_name, tool_call_args in tool_calls:
+                tool_instance = tool_instances.get(tool_call_name)
+                if not tool_instance:
+                    tool_response = {
+                        "tool_call_id": tool_call_id,
+                        "tool_call_name": tool_call_name,
+                        "tool_response": f"there is not a tool named {tool_call_name}",
+                        "meta": ToolInvokeMeta.error_instance(f"there is not a tool named {tool_call_name}").to_dict(),
+                    }
+                else:
+                    # invoke tool
+                    tool_invoke_response, message_files, tool_invoke_meta = ToolEngine.agent_invoke(
+                        tool=tool_instance,
+                        tool_parameters=tool_call_args,
+                        user_id=self.user_id,
+                        tenant_id=self.tenant_id,
+                        message=self.message,
+                        invoke_from=self.application_generate_entity.invoke_from,
+                        agent_tool_callback=self.agent_callback,
+                        trace_manager=trace_manager,
+                        app_id=self.application_generate_entity.app_config.app_id,
+                        message_id=self.message.id,
+                        conversation_id=self.conversation.id,
+                    )
+                    # publish files
+                    for message_file_id in message_files:
+                        # publish message file
+                        self.queue_manager.publish(
+                            QueueMessageFileEvent(message_file_id=message_file_id), PublishFrom.APPLICATION_MANAGER
+                        )
+                        # add message file ids
+                        message_file_ids.append(message_file_id)
+
+                    tool_response = {
+                        "tool_call_id": tool_call_id,
+                        "tool_call_name": tool_call_name,
+                        "tool_response": tool_invoke_response,
+                        "meta": tool_invoke_meta.to_dict(),
+                    }
+
+                tool_responses.append(tool_response)
+                if tool_response["tool_response"] is not None:
+                    self._current_thoughts.append(
+                        ToolPromptMessage(
+                            content=str(tool_response["tool_response"]),
+                            tool_call_id=tool_call_id,
+                            name=tool_call_name,
+                        )
+                    )
+
+            if len(tool_responses) > 0:
+                # save agent thought
+                self.save_agent_thought(
+                    agent_thought_id=agent_thought_id,
+                    tool_name="",
+                    tool_input="",
+                    thought="",
+                    tool_invoke_meta={
+                        tool_response["tool_call_name"]: tool_response["meta"] for tool_response in tool_responses
+                    },
+                    observation={
+                        tool_response["tool_call_name"]: tool_response["tool_response"]
+                        for tool_response in tool_responses
+                    },
+                    answer="",
+                    messages_ids=message_file_ids,
+                )
+                self.queue_manager.publish(
+                    QueueAgentThoughtEvent(agent_thought_id=agent_thought_id), PublishFrom.APPLICATION_MANAGER
+                )
+
+            # update prompt tool
+            for prompt_tool in prompt_messages_tools:
+                self.update_prompt_message_tool(tool_instances[prompt_tool.name], prompt_tool)
+
+            iteration_step += 1
+
+        # publish end event
+        self.queue_manager.publish(
+            QueueMessageEndEvent(
+                llm_result=LLMResult(
+                    model=model_instance.model,
+                    prompt_messages=prompt_messages,
+                    message=AssistantPromptMessage(content=final_answer),
+                    usage=llm_usage["usage"] or LLMUsage.empty_usage(),
+                    system_fingerprint="",
+                )
+            ),
+            PublishFrom.APPLICATION_MANAGER,
+        )
+
+    def check_tool_calls(self, llm_result_chunk: LLMResultChunk) -> bool:
+        """
+        Check if there is any tool call in llm result chunk
+        """
+        if llm_result_chunk.delta.message.tool_calls:
+            return True
+        return False
+
+    def check_blocking_tool_calls(self, llm_result: LLMResult) -> bool:
+        """
+        Check if there is any blocking tool call in llm result
+        """
+        if llm_result.message.tool_calls:
+            return True
+        return False
+
+    def extract_tool_calls(self, llm_result_chunk: LLMResultChunk) -> list[tuple[str, str, dict[str, Any]]]:
+        """
+        Extract tool calls from llm result chunk
+
+        Returns:
+            List[Tuple[str, str, Dict[str, Any]]]: [(tool_call_id, tool_call_name, tool_call_args)]
+        """
+        tool_calls = []
+        for prompt_message in llm_result_chunk.delta.message.tool_calls:
+            args = {}
+            if prompt_message.function.arguments != "":
+                args = json.loads(prompt_message.function.arguments)
+
+            tool_calls.append(
+                (
+                    prompt_message.id,
+                    prompt_message.function.name,
+                    args,
+                )
+            )
+
+        return tool_calls
+
+    def extract_blocking_tool_calls(self, llm_result: LLMResult) -> list[tuple[str, str, dict[str, Any]]]:
+        """
+        Extract blocking tool calls from llm result
+
+        Returns:
+            List[Tuple[str, str, Dict[str, Any]]]: [(tool_call_id, tool_call_name, tool_call_args)]
+        """
+        tool_calls = []
+        for prompt_message in llm_result.message.tool_calls:
+            args = {}
+            if prompt_message.function.arguments != "":
+                args = json.loads(prompt_message.function.arguments)
+
+            tool_calls.append(
+                (
+                    prompt_message.id,
+                    prompt_message.function.name,
+                    args,
+                )
+            )
+
+        return tool_calls
+
+    def _init_system_message(self, prompt_template: str, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
+        """
+        Initialize system message
+        """
+        if not prompt_messages and prompt_template:
+            return [
+                SystemPromptMessage(content=prompt_template),
+            ]
+
+        if prompt_messages and not isinstance(prompt_messages[0], SystemPromptMessage) and prompt_template:
+            prompt_messages.insert(0, SystemPromptMessage(content=prompt_template))
+
+        return prompt_messages or []
+
+    def _organize_user_query(self, query: str, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
+        """
+        Organize user query
+        """
+        if self.files:
+            # get image detail config
+            image_detail_config = (
+                self.application_generate_entity.file_upload_config.image_config.detail
+                if (
+                    self.application_generate_entity.file_upload_config
+                    and self.application_generate_entity.file_upload_config.image_config
+                )
+                else None
+            )
+            image_detail_config = image_detail_config or ImagePromptMessageContent.DETAIL.LOW
+
+            prompt_message_contents: list[PromptMessageContentUnionTypes] = []
+            for file in self.files:
+                prompt_message_contents.append(
+                    file_manager.to_prompt_message_content(
+                        file,
+                        image_detail_config=image_detail_config,
+                    )
+                )
+            prompt_message_contents.append(TextPromptMessageContent(data=query))
+
+            prompt_messages.append(UserPromptMessage(content=prompt_message_contents))
+        else:
+            prompt_messages.append(UserPromptMessage(content=query))
+
+        return prompt_messages
+
+    def _clear_user_prompt_image_messages(self, prompt_messages: list[PromptMessage]) -> list[PromptMessage]:
+        """
+        As for now, gpt supports both fc and vision at the first iteration.
+        We need to remove the image messages from the prompt messages at the first iteration.
+        """
+        prompt_messages = deepcopy(prompt_messages)
+
+        for prompt_message in prompt_messages:
+            if isinstance(prompt_message, UserPromptMessage):
+                if isinstance(prompt_message.content, list):
+                    prompt_message.content = "\n".join(
+                        [
+                            content.data
+                            if content.type == PromptMessageContentType.TEXT
+                            else "[image]"
+                            if content.type == PromptMessageContentType.IMAGE
+                            else "[file]"
+                            for content in prompt_message.content
+                        ]
+                    )
+
+        return prompt_messages
+
+    def _organize_prompt_messages(self):
+        prompt_template = self.app_config.prompt_template.simple_prompt_template or ""
+        self.history_prompt_messages = self._init_system_message(prompt_template, self.history_prompt_messages)
+        query_prompt_messages = self._organize_user_query(self.query or "", [])
+
+        self.history_prompt_messages = AgentHistoryPromptTransform(
+            model_config=self.model_config,
+            prompt_messages=[*query_prompt_messages, *self._current_thoughts],
+            history_messages=self.history_prompt_messages,
+            memory=self.memory,
+        ).get_prompt()
+
+        prompt_messages = [*self.history_prompt_messages, *query_prompt_messages, *self._current_thoughts]
+        if len(self._current_thoughts) != 0:
+            # clear messages after the first iteration
+            prompt_messages = self._clear_user_prompt_image_messages(prompt_messages)
+        return prompt_messages
@@ -1,55 +0,0 @@
-# Agent Patterns
-
-A unified agent pattern module that powers both Agent V2 workflow nodes and agent applications. Strategies share a common execution contract while adapting to model capabilities and tool availability.
-
-## Overview
-
-The module applies a strategy pattern around LLM/tool orchestration. `StrategyFactory` auto-selects the best implementation based on model features or an explicit agent strategy, and each strategy streams logs and usage consistently.
-
-## Key Features
-
- **Dual strategies**
-  - `FunctionCallStrategy`: uses native LLM function/tool calling when the model exposes `TOOL_CALL`, `MULTI_TOOL_CALL`, or `STREAM_TOOL_CALL`.
-  - `ReActStrategy`: ReAct (reasoning + acting) flow driven by `CotAgentOutputParser`, used when function calling is unavailable or explicitly requested.
- **Explicit or auto selection**
-  - `StrategyFactory.create_strategy` prefers an explicit `AgentEntity.Strategy` (FUNCTION_CALLING or CHAIN_OF_THOUGHT).
-  - Otherwise it falls back to function calling when tool-call features exist, or ReAct when they do not.
- **Unified execution contract**
-  - `AgentPattern.run` yields streaming `AgentLog` entries and `LLMResultChunk` data, returning an `AgentResult` with text, files, usage, and `finish_reason`.
-  - Iterations are configurable and hard-capped at 99 rounds; the last round forces a final answer by withholding tools.
- **Tool handling and hooks**
-  - Tools convert to `PromptMessageTool` objects before invocation.
-  - Optional `tool_invoke_hook` lets callers override tool execution (e.g., agent apps) while workflow runs use `ToolEngine.generic_invoke`.
-  - Tool outputs support text, links, JSON, variables, blobs, retriever resources, and file attachments; `target=="self"` files are reloaded into model context, others are returned as outputs.
- **File-aware arguments**
-  - Tool args accept `[File: <id>]` or `[Files: <id1, id2>]` placeholders that resolve to `File` objects before invocation, enabling models to reference uploaded files safely.
- **ReAct prompt shaping**
-  - System prompts replace `{{instruction}}`, `{{tools}}`, and `{{tool_names}}` placeholders.
-  - Adds `Observation` to stop sequences and appends scratchpad text so the model sees prior Thought/Action/Observation history.
- **Observability and accounting**
-  - Standardized `AgentLog` entries for rounds, model thoughts, and tool calls, including usage aggregation (`LLMUsage`) across streaming and non-streaming paths.
-
-## Architecture
-
-```
-agent/patterns/
-├── base.py              # Shared utilities: logging, usage, tool invocation, file handling
-├── function_call.py     # Native function-calling loop with tool execution
-├── react.py             # ReAct loop with CoT parsing and scratchpad wiring
-└── strategy_factory.py  # Strategy selection by model features or explicit override
-```
-
-## Usage
-
- For auto-selection:
-  - Call `StrategyFactory.create_strategy(model_features, model_instance, context, tools, files, ...)` and run the returned strategy with prompt messages and model params.
- For explicit behavior:
-  - Pass `agent_strategy=AgentEntity.Strategy.FUNCTION_CALLING` to force native calls (falls back to ReAct if unsupported), or `CHAIN_OF_THOUGHT` to force ReAct.
- Both strategies stream chunks and logs; collect the generator output until it returns an `AgentResult`.
-
-## Integration Points
-
- **Model runtime**: delegates to `ModelInstance.invoke_llm` for both streaming and non-streaming calls.
- **Tool system**: defaults to `ToolEngine.generic_invoke`, with `tool_invoke_hook` for custom callers.
- **Files**: flows through `File` objects for tool inputs/outputs and model-context attachments.
- **Execution context**: `ExecutionContext` fields (user/app/conversation/message) propagate to tool invocations and logging.
@@ -1,19 +0,0 @@
-"""Agent patterns module.
-
-This module provides different strategies for agent execution:
- FunctionCallStrategy: Uses native function/tool calling
- ReActStrategy: Uses ReAct (Reasoning + Acting) approach
- StrategyFactory: Factory for creating strategies based on model features
-"""
-
-from .base import AgentPattern
-from .function_call import FunctionCallStrategy
-from .react import ReActStrategy
-from .strategy_factory import StrategyFactory
-
-__all__ = [
-    "AgentPattern",
-    "FunctionCallStrategy",
-    "ReActStrategy",
-    "StrategyFactory",
-]
@@ -1,474 +0,0 @@
-"""Base class for agent strategies."""
-
-from __future__ import annotations
-
-import json
-import re
-import time
-from abc import ABC, abstractmethod
-from collections.abc import Callable, Generator
-from typing import TYPE_CHECKING, Any
-
-from core.agent.entities import AgentLog, AgentResult, ExecutionContext
-from core.file import File
-from core.model_manager import ModelInstance
-from core.model_runtime.entities import (
-    AssistantPromptMessage,
-    LLMResult,
-    LLMResultChunk,
-    LLMResultChunkDelta,
-    PromptMessage,
-    PromptMessageTool,
-)
-from core.model_runtime.entities.llm_entities import LLMUsage
-from core.model_runtime.entities.message_entities import TextPromptMessageContent
-from core.tools.entities.tool_entities import ToolInvokeMessage, ToolInvokeMeta
-
-if TYPE_CHECKING:
-    from core.tools.__base.tool import Tool
-
-# Type alias for tool invoke hook
-# Returns: (response_content, message_file_ids, tool_invoke_meta)
-ToolInvokeHook = Callable[["Tool", dict[str, Any], str], tuple[str, list[str], ToolInvokeMeta]]
-
-
-class AgentPattern(ABC):
-    """Base class for agent execution strategies."""
-
-    def __init__(
-        self,
-        model_instance: ModelInstance,
-        tools: list[Tool],
-        context: ExecutionContext,
-        max_iterations: int = 10,
-        workflow_call_depth: int = 0,
-        files: list[File] = [],
-        tool_invoke_hook: ToolInvokeHook | None = None,
-    ):
-        """Initialize the agent strategy."""
-        self.model_instance = model_instance
-        self.tools = tools
-        self.context = context
-        self.max_iterations = min(max_iterations, 99)  # Cap at 99 iterations
-        self.workflow_call_depth = workflow_call_depth
-        self.files: list[File] = files
-        self.tool_invoke_hook = tool_invoke_hook
-
-    @abstractmethod
-    def run(
-        self,
-        prompt_messages: list[PromptMessage],
-        model_parameters: dict[str, Any],
-        stop: list[str] = [],
-        stream: bool = True,
-    ) -> Generator[LLMResultChunk | AgentLog, None, AgentResult]:
-        """Execute the agent strategy."""
-        pass
-
-    def _accumulate_usage(self, total_usage: dict[str, Any], delta_usage: LLMUsage) -> None:
-        """Accumulate LLM usage statistics."""
-        if not total_usage.get("usage"):
-            # Create a copy to avoid modifying the original
-            total_usage["usage"] = LLMUsage(
-                prompt_tokens=delta_usage.prompt_tokens,
-                prompt_unit_price=delta_usage.prompt_unit_price,
-                prompt_price_unit=delta_usage.prompt_price_unit,
-                prompt_price=delta_usage.prompt_price,
-                completion_tokens=delta_usage.completion_tokens,
-                completion_unit_price=delta_usage.completion_unit_price,
-                completion_price_unit=delta_usage.completion_price_unit,
-                completion_price=delta_usage.completion_price,
-                total_tokens=delta_usage.total_tokens,
-                total_price=delta_usage.total_price,
-                currency=delta_usage.currency,
-                latency=delta_usage.latency,
-            )
-        else:
-            current: LLMUsage = total_usage["usage"]
-            current.prompt_tokens += delta_usage.prompt_tokens
-            current.completion_tokens += delta_usage.completion_tokens
-            current.total_tokens += delta_usage.total_tokens
-            current.prompt_price += delta_usage.prompt_price
-            current.completion_price += delta_usage.completion_price
-            current.total_price += delta_usage.total_price
-
-    def _extract_content(self, content: Any) -> str:
-        """Extract text content from message content."""
-        if isinstance(content, list):
-            # Content items are PromptMessageContentUnionTypes
-            text_parts = []
-            for c in content:
-                # Check if it's a TextPromptMessageContent (which has data attribute)
-                if isinstance(c, TextPromptMessageContent):
-                    text_parts.append(c.data)
-            return "".join(text_parts)
-        return str(content)
-
-    def _has_tool_calls(self, chunk: LLMResultChunk) -> bool:
-        """Check if chunk contains tool calls."""
-        # LLMResultChunk always has delta attribute
-        return bool(chunk.delta.message and chunk.delta.message.tool_calls)
-
-    def _has_tool_calls_result(self, result: LLMResult) -> bool:
-        """Check if result contains tool calls (non-streaming)."""
-        # LLMResult always has message attribute
-        return bool(result.message and result.message.tool_calls)
-
-    def _extract_tool_calls(self, chunk: LLMResultChunk) -> list[tuple[str, str, dict[str, Any]]]:
-        """Extract tool calls from streaming chunk."""
-        tool_calls: list[tuple[str, str, dict[str, Any]]] = []
-        if chunk.delta.message and chunk.delta.message.tool_calls:
-            for tool_call in chunk.delta.message.tool_calls:
-                if tool_call.function:
-                    try:
-                        args = json.loads(tool_call.function.arguments) if tool_call.function.arguments else {}
-                    except json.JSONDecodeError:
-                        args = {}
-                    tool_calls.append((tool_call.id or "", tool_call.function.name, args))
-        return tool_calls
-
-    def _extract_tool_calls_result(self, result: LLMResult) -> list[tuple[str, str, dict[str, Any]]]:
-        """Extract tool calls from non-streaming result."""
-        tool_calls = []
-        if result.message and result.message.tool_calls:
-            for tool_call in result.message.tool_calls:
-                if tool_call.function:
-                    try:
-                        args = json.loads(tool_call.function.arguments) if tool_call.function.arguments else {}
-                    except json.JSONDecodeError:
-                        args = {}
-                    tool_calls.append((tool_call.id or "", tool_call.function.name, args))
-        return tool_calls
-
-    def _extract_text_from_message(self, message: PromptMessage) -> str:
-        """Extract text content from a prompt message."""
-        # PromptMessage always has content attribute
-        content = message.content
-        if isinstance(content, str):
-            return content
-        elif isinstance(content, list):
-            # Extract text from content list
-            text_parts = []
-            for item in content:
-                if isinstance(item, TextPromptMessageContent):
-                    text_parts.append(item.data)
-            return " ".join(text_parts)
-        return ""
-
-    def _get_tool_metadata(self, tool_instance: Tool) -> dict[AgentLog.LogMetadata, Any]:
-        """Get metadata for a tool including provider and icon info."""
-        from core.tools.tool_manager import ToolManager
-
-        metadata: dict[AgentLog.LogMetadata, Any] = {}
-        if tool_instance.entity and tool_instance.entity.identity:
-            identity = tool_instance.entity.identity
-            if identity.provider:
-                metadata[AgentLog.LogMetadata.PROVIDER] = identity.provider
-
-            # Get icon using ToolManager for proper URL generation
-            tenant_id = self.context.tenant_id
-            if tenant_id and identity.provider:
-                try:
-                    provider_type = tool_instance.tool_provider_type()
-                    icon = ToolManager.get_tool_icon(tenant_id, provider_type, identity.provider)
-                    if isinstance(icon, str):
-                        metadata[AgentLog.LogMetadata.ICON] = icon
-                    elif isinstance(icon, dict):
-                        # Handle icon dict with background/content or light/dark variants
-                        metadata[AgentLog.LogMetadata.ICON] = icon
-                except Exception:
-                    # Fallback to identity.icon if ToolManager fails
-                    if identity.icon:
-                        metadata[AgentLog.LogMetadata.ICON] = identity.icon
-            elif identity.icon:
-                metadata[AgentLog.LogMetadata.ICON] = identity.icon
-        return metadata
-
-    def _create_log(
-        self,
-        label: str,
-        log_type: AgentLog.LogType,
-        status: AgentLog.LogStatus,
-        data: dict[str, Any] | None = None,
-        parent_id: str | None = None,
-        extra_metadata: dict[AgentLog.LogMetadata, Any] | None = None,
-    ) -> AgentLog:
-        """Create a new AgentLog with standard metadata."""
-        metadata: dict[AgentLog.LogMetadata, Any] = {
-            AgentLog.LogMetadata.STARTED_AT: time.perf_counter(),
-        }
-        if extra_metadata:
-            metadata.update(extra_metadata)
-
-        return AgentLog(
-            label=label,
-            log_type=log_type,
-            status=status,
-            data=data or {},
-            parent_id=parent_id,
-            metadata=metadata,
-        )
-
-    def _finish_log(
-        self,
-        log: AgentLog,
-        data: dict[str, Any] | None = None,
-        usage: LLMUsage | None = None,
-    ) -> AgentLog:
-        """Finish an AgentLog by updating its status and metadata."""
-        log.status = AgentLog.LogStatus.SUCCESS
-
-        if data is not None:
-            log.data = data
-
-        # Calculate elapsed time
-        started_at = log.metadata.get(AgentLog.LogMetadata.STARTED_AT, time.perf_counter())
-        finished_at = time.perf_counter()
-
-        # Update metadata
-        log.metadata = {
-            **log.metadata,
-            AgentLog.LogMetadata.FINISHED_AT: finished_at,
-            # Calculate elapsed time in seconds
-            AgentLog.LogMetadata.ELAPSED_TIME: round(finished_at - started_at, 4),
-        }
-
-        # Add usage information if provided
-        if usage:
-            log.metadata.update(
-                {
-                    AgentLog.LogMetadata.TOTAL_PRICE: usage.total_price,
-                    AgentLog.LogMetadata.CURRENCY: usage.currency,
-                    AgentLog.LogMetadata.TOTAL_TOKENS: usage.total_tokens,
-                    AgentLog.LogMetadata.LLM_USAGE: usage,
-                }
-            )
-
-        return log
-
-    def _replace_file_references(self, tool_args: dict[str, Any]) -> dict[str, Any]:
-        """
-        Replace file references in tool arguments with actual File objects.
-
-        Args:
-            tool_args: Dictionary of tool arguments
-
-        Returns:
-            Updated tool arguments with file references replaced
-        """
-        # Process each argument in the dictionary
-        processed_args: dict[str, Any] = {}
-        for key, value in tool_args.items():
-            processed_args[key] = self._process_file_reference(value)
-        return processed_args
-
-    def _process_file_reference(self, data: Any) -> Any:
-        """
-        Recursively process data to replace file references.
-        Supports both single file [File: file_id] and multiple files [Files: file_id1, file_id2, ...].
-
-        Args:
-            data: The data to process (can be dict, list, str, or other types)
-
-        Returns:
-            Processed data with file references replaced
-        """
-        single_file_pattern = re.compile(r"^\[File:\s*([^\]]+)\]$")
-        multiple_files_pattern = re.compile(r"^\[Files:\s*([^\]]+)\]$")
-
-        if isinstance(data, dict):
-            # Process dictionary recursively
-            return {key: self._process_file_reference(value) for key, value in data.items()}
-        elif isinstance(data, list):
-            # Process list recursively
-            return [self._process_file_reference(item) for item in data]
-        elif isinstance(data, str):
-            # Check for single file pattern [File: file_id]
-            single_match = single_file_pattern.match(data.strip())
-            if single_match:
-                file_id = single_match.group(1).strip()
-                # Find the file in self.files
-                for file in self.files:
-                    if file.id and str(file.id) == file_id:
-                        return file
-                # If file not found, return original value
-                return data
-
-            # Check for multiple files pattern [Files: file_id1, file_id2, ...]
-            multiple_match = multiple_files_pattern.match(data.strip())
-            if multiple_match:
-                file_ids_str = multiple_match.group(1).strip()
-                # Split by comma and strip whitespace
-                file_ids = [fid.strip() for fid in file_ids_str.split(",")]
-
-                # Find all matching files
-                matched_files: list[File] = []
-                for file_id in file_ids:
-                    for file in self.files:
-                        if file.id and str(file.id) == file_id:
-                            matched_files.append(file)
-                            break
-
-                # Return list of files if any were found, otherwise return original
-                return matched_files or data
-
-            return data
-        else:
-            # Return other types as-is
-            return data
-
-    def _create_text_chunk(self, text: str, prompt_messages: list[PromptMessage]) -> LLMResultChunk:
-        """Create a text chunk for streaming."""
-        return LLMResultChunk(
-            model=self.model_instance.model,
-            prompt_messages=prompt_messages,
-            delta=LLMResultChunkDelta(
-                index=0,
-                message=AssistantPromptMessage(content=text),
-                usage=None,
-            ),
-            system_fingerprint="",
-        )
-
-    def _invoke_tool(
-        self,
-        tool_instance: Tool,
-        tool_args: dict[str, Any],
-        tool_name: str,
-    ) -> tuple[str, list[File], ToolInvokeMeta | None]:
-        """
-        Invoke a tool and collect its response.
-
-        Args:
-            tool_instance: The tool instance to invoke
-            tool_args: Tool arguments
-            tool_name: Name of the tool
-
-        Returns:
-            Tuple of (response_content, tool_files, tool_invoke_meta)
-        """
-        # Process tool_args to replace file references with actual File objects
-        tool_args = self._replace_file_references(tool_args)
-
-        # If a tool invoke hook is set, use it instead of generic_invoke
-        if self.tool_invoke_hook:
-            response_content, _, tool_invoke_meta = self.tool_invoke_hook(tool_instance, tool_args, tool_name)
-            # Note: message_file_ids are stored in DB, we don't convert them to File objects here
-            # The caller (AgentAppRunner) handles file publishing
-            return response_content, [], tool_invoke_meta
-
-        # Default: use generic_invoke for workflow scenarios
-        # Import here to avoid circular import
-        from core.tools.tool_engine import DifyWorkflowCallbackHandler, ToolEngine
-
-        tool_response = ToolEngine().generic_invoke(
-            tool=tool_instance,
-            tool_parameters=tool_args,
-            user_id=self.context.user_id or "",
-            workflow_tool_callback=DifyWorkflowCallbackHandler(),
-            workflow_call_depth=self.workflow_call_depth,
-            app_id=self.context.app_id,
-            conversation_id=self.context.conversation_id,
-            message_id=self.context.message_id,
-        )
-
-        # Collect response and files
-        response_content = ""
-        tool_files: list[File] = []
-
-        for response in tool_response:
-            if response.type == ToolInvokeMessage.MessageType.TEXT:
-                assert isinstance(response.message, ToolInvokeMessage.TextMessage)
-                response_content += response.message.text
-
-            elif response.type == ToolInvokeMessage.MessageType.LINK:
-                # Handle link messages
-                if isinstance(response.message, ToolInvokeMessage.TextMessage):
-                    response_content += f"[Link: {response.message.text}]"
-
-            elif response.type == ToolInvokeMessage.MessageType.IMAGE:
-                # Handle image URL messages
-                if isinstance(response.message, ToolInvokeMessage.TextMessage):
-                    response_content += f"[Image: {response.message.text}]"
-
-            elif response.type == ToolInvokeMessage.MessageType.IMAGE_LINK:
-                # Handle image link messages
-                if isinstance(response.message, ToolInvokeMessage.TextMessage):
-                    response_content += f"[Image: {response.message.text}]"
-
-            elif response.type == ToolInvokeMessage.MessageType.BINARY_LINK:
-                # Handle binary file link messages
-                if isinstance(response.message, ToolInvokeMessage.TextMessage):
-                    filename = response.meta.get("filename", "file") if response.meta else "file"
-                    response_content += f"[File: {filename} - {response.message.text}]"
-
-            elif response.type == ToolInvokeMessage.MessageType.JSON:
-                # Handle JSON messages
-                if isinstance(response.message, ToolInvokeMessage.JsonMessage):
-                    response_content += json.dumps(response.message.json_object, ensure_ascii=False, indent=2)
-
-            elif response.type == ToolInvokeMessage.MessageType.BLOB:
-                # Handle blob messages - convert to text representation
-                if isinstance(response.message, ToolInvokeMessage.BlobMessage):
-                    mime_type = (
-                        response.meta.get("mime_type", "application/octet-stream")
-                        if response.meta
-                        else "application/octet-stream"
-                    )
-                    size = len(response.message.blob)
-                    response_content += f"[Binary data: {mime_type}, size: {size} bytes]"
-
-            elif response.type == ToolInvokeMessage.MessageType.VARIABLE:
-                # Handle variable messages
-                if isinstance(response.message, ToolInvokeMessage.VariableMessage):
-                    var_name = response.message.variable_name
-                    var_value = response.message.variable_value
-                    if isinstance(var_value, str):
-                        response_content += var_value
-                    else:
-                        response_content += f"[Variable {var_name}: {json.dumps(var_value, ensure_ascii=False)}]"
-
-            elif response.type == ToolInvokeMessage.MessageType.BLOB_CHUNK:
-                # Handle blob chunk messages - these are parts of a larger blob
-                if isinstance(response.message, ToolInvokeMessage.BlobChunkMessage):
-                    response_content += f"[Blob chunk {response.message.sequence}: {len(response.message.blob)} bytes]"
-
-            elif response.type == ToolInvokeMessage.MessageType.RETRIEVER_RESOURCES:
-                # Handle retriever resources messages
-                if isinstance(response.message, ToolInvokeMessage.RetrieverResourceMessage):
-                    response_content += response.message.context
-
-            elif response.type == ToolInvokeMessage.MessageType.FILE:
-                # Extract file from meta
-                if response.meta and "file" in response.meta:
-                    file = response.meta["file"]
-                    if isinstance(file, File):
-                        # Check if file is for model or tool output
-                        if response.meta.get("target") == "self":
-                            # File is for model - add to files for next prompt
-                            self.files.append(file)
-                            response_content += f"File '{file.filename}' has been loaded into your context."
-                        else:
-                            # File is tool output
-                            tool_files.append(file)
-
-        return response_content, tool_files, None
-
-    def _find_tool_by_name(self, tool_name: str) -> Tool | None:
-        """Find a tool instance by its name."""
-        for tool in self.tools:
-            if tool.entity.identity.name == tool_name:
-                return tool
-        return None
-
-    def _convert_tools_to_prompt_format(self) -> list[PromptMessageTool]:
-        """Convert tools to prompt message format."""
-        prompt_tools: list[PromptMessageTool] = []
-        for tool in self.tools:
-            prompt_tools.append(tool.to_prompt_message_tool())
-        return prompt_tools
-
-    def _update_usage_with_empty(self, llm_usage: dict[str, Any]) -> None:
-        """Initialize usage tracking with empty usage if not set."""
-        if "usage" not in llm_usage or llm_usage["usage"] is None:
-            llm_usage["usage"] = LLMUsage.empty_usage()
@@ -1,299 +0,0 @@
-"""Function Call strategy implementation."""
-
-import json
-from collections.abc import Generator
-from typing import Any, Union
-
-from core.agent.entities import AgentLog, AgentResult
-from core.file import File
-from core.model_runtime.entities import (
-    AssistantPromptMessage,
-    LLMResult,
-    LLMResultChunk,
-    LLMResultChunkDelta,
-    LLMUsage,
-    PromptMessage,
-    PromptMessageTool,
-    ToolPromptMessage,
-)
-from core.tools.entities.tool_entities import ToolInvokeMeta
-
-from .base import AgentPattern
-
-
-class FunctionCallStrategy(AgentPattern):
-    """Function Call strategy using model's native tool calling capability."""
-
-    def run(
-        self,
-        prompt_messages: list[PromptMessage],
-        model_parameters: dict[str, Any],
-        stop: list[str] = [],
-        stream: bool = True,
-    ) -> Generator[LLMResultChunk | AgentLog, None, AgentResult]:
-        """Execute the function call agent strategy."""
-        # Convert tools to prompt format
-        prompt_tools: list[PromptMessageTool] = self._convert_tools_to_prompt_format()
-
-        # Initialize tracking
-        iteration_step: int = 1
-        max_iterations: int = self.max_iterations + 1
-        function_call_state: bool = True
-        total_usage: dict[str, LLMUsage | None] = {"usage": None}
-        messages: list[PromptMessage] = list(prompt_messages)  # Create mutable copy
-        final_text: str = ""
-        finish_reason: str | None = None
-        output_files: list[File] = []  # Track files produced by tools
-
-        while function_call_state and iteration_step <= max_iterations:
-            function_call_state = False
-            round_log = self._create_log(
-                label=f"ROUND {iteration_step}",
-                log_type=AgentLog.LogType.ROUND,
-                status=AgentLog.LogStatus.START,
-                data={},
-            )
-            yield round_log
-            # On last iteration, remove tools to force final answer
-            current_tools: list[PromptMessageTool] = [] if iteration_step == max_iterations else prompt_tools
-            model_log = self._create_log(
-                label=f"{self.model_instance.model} Thought",
-                log_type=AgentLog.LogType.THOUGHT,
-                status=AgentLog.LogStatus.START,
-                data={},
-                parent_id=round_log.id,
-                extra_metadata={
-                    AgentLog.LogMetadata.PROVIDER: self.model_instance.provider,
-                },
-            )
-            yield model_log
-
-            # Track usage for this round only
-            round_usage: dict[str, LLMUsage | None] = {"usage": None}
-
-            # Invoke model
-            chunks: Union[Generator[LLMResultChunk, None, None], LLMResult] = self.model_instance.invoke_llm(
-                prompt_messages=messages,
-                model_parameters=model_parameters,
-                tools=current_tools,
-                stop=stop,
-                stream=stream,
-                user=self.context.user_id,
-                callbacks=[],
-            )
-
-            # Process response
-            tool_calls, response_content, chunk_finish_reason = yield from self._handle_chunks(
-                chunks, round_usage, model_log
-            )
-            messages.append(self._create_assistant_message(response_content, tool_calls))
-
-            # Accumulate to total usage
-            round_usage_value = round_usage.get("usage")
-            if round_usage_value:
-                self._accumulate_usage(total_usage, round_usage_value)
-
-            # Update final text if no tool calls (this is likely the final answer)
-            if not tool_calls:
-                final_text = response_content
-
-            # Update finish reason
-            if chunk_finish_reason:
-                finish_reason = chunk_finish_reason
-
-            # Process tool calls
-            tool_outputs: dict[str, str] = {}
-            if tool_calls:
-                function_call_state = True
-                # Execute tools
-                for tool_call_id, tool_name, tool_args in tool_calls:
-                    tool_response, tool_files, _ = yield from self._handle_tool_call(
-                        tool_name, tool_args, tool_call_id, messages, round_log
-                    )
-                    tool_outputs[tool_name] = tool_response
-                    # Track files produced by tools
-                    output_files.extend(tool_files)
-            yield self._finish_log(
-                round_log,
-                data={
-                    "llm_result": response_content,
-                    "tool_calls": [
-                        {"name": tc[1], "args": tc[2], "output": tool_outputs.get(tc[1], "")} for tc in tool_calls
-                    ]
-                    if tool_calls
-                    else [],
-                    "final_answer": final_text if not function_call_state else None,
-                },
-                usage=round_usage.get("usage"),
-            )
-            iteration_step += 1
-
-        # Return final result
-        from core.agent.entities import AgentResult
-
-        return AgentResult(
-            text=final_text,
-            files=output_files,
-            usage=total_usage.get("usage") or LLMUsage.empty_usage(),
-            finish_reason=finish_reason,
-        )
-
-    def _handle_chunks(
-        self,
-        chunks: Union[Generator[LLMResultChunk, None, None], LLMResult],
-        llm_usage: dict[str, LLMUsage | None],
-        start_log: AgentLog,
-    ) -> Generator[
-        LLMResultChunk | AgentLog,
-        None,
-        tuple[list[tuple[str, str, dict[str, Any]]], str, str | None],
-    ]:
-        """Handle LLM response chunks and extract tool calls and content.
-
-        Returns a tuple of (tool_calls, response_content, finish_reason).
-        """
-        tool_calls: list[tuple[str, str, dict[str, Any]]] = []
-        response_content: str = ""
-        finish_reason: str | None = None
-        if isinstance(chunks, Generator):
-            # Streaming response
-            for chunk in chunks:
-                # Extract tool calls
-                if self._has_tool_calls(chunk):
-                    tool_calls.extend(self._extract_tool_calls(chunk))
-
-                # Extract content
-                if chunk.delta.message and chunk.delta.message.content:
-                    response_content += self._extract_content(chunk.delta.message.content)
-
-                # Track usage
-                if chunk.delta.usage:
-                    self._accumulate_usage(llm_usage, chunk.delta.usage)
-
-                # Capture finish reason
-                if chunk.delta.finish_reason:
-                    finish_reason = chunk.delta.finish_reason
-
-                yield chunk
-        else:
-            # Non-streaming response
-            result: LLMResult = chunks
-
-            if self._has_tool_calls_result(result):
-                tool_calls.extend(self._extract_tool_calls_result(result))
-
-            if result.message and result.message.content:
-                response_content += self._extract_content(result.message.content)
-
-            if result.usage:
-                self._accumulate_usage(llm_usage, result.usage)
-
-            # Convert to streaming format
-            yield LLMResultChunk(
-                model=result.model,
-                prompt_messages=result.prompt_messages,
-                delta=LLMResultChunkDelta(index=0, message=result.message, usage=result.usage),
-            )
-        yield self._finish_log(
-            start_log,
-            data={
-                "result": response_content,
-            },
-            usage=llm_usage.get("usage"),
-        )
-        return tool_calls, response_content, finish_reason
-
-    def _create_assistant_message(
-        self, content: str, tool_calls: list[tuple[str, str, dict[str, Any]]] | None = None
-    ) -> AssistantPromptMessage:
-        """Create assistant message with tool calls."""
-        if tool_calls is None:
-            return AssistantPromptMessage(content=content)
-        return AssistantPromptMessage(
-            content=content or "",
-            tool_calls=[
-                AssistantPromptMessage.ToolCall(
-                    id=tc[0],
-                    type="function",
-                    function=AssistantPromptMessage.ToolCall.ToolCallFunction(name=tc[1], arguments=json.dumps(tc[2])),
-                )
-                for tc in tool_calls
-            ],
-        )
-
-    def _handle_tool_call(
-        self,
-        tool_name: str,
-        tool_args: dict[str, Any],
-        tool_call_id: str,
-        messages: list[PromptMessage],
-        round_log: AgentLog,
-    ) -> Generator[AgentLog, None, tuple[str, list[File], ToolInvokeMeta | None]]:
-        """Handle a single tool call and return response with files and meta."""
-        # Find tool
-        tool_instance = self._find_tool_by_name(tool_name)
-        if not tool_instance:
-            raise ValueError(f"Tool {tool_name} not found")
-
-        # Get tool metadata (provider, icon, etc.)
-        tool_metadata = self._get_tool_metadata(tool_instance)
-
-        # Create tool call log
-        tool_call_log = self._create_log(
-            label=f"CALL {tool_name}",
-            log_type=AgentLog.LogType.TOOL_CALL,
-            status=AgentLog.LogStatus.START,
-            data={
-                "tool_call_id": tool_call_id,
-                "tool_name": tool_name,
-                "tool_args": tool_args,
-            },
-            parent_id=round_log.id,
-            extra_metadata=tool_metadata,
-        )
-        yield tool_call_log
-
-        # Invoke tool using base class method with error handling
-        try:
-            response_content, tool_files, tool_invoke_meta = self._invoke_tool(tool_instance, tool_args, tool_name)
-
-            yield self._finish_log(
-                tool_call_log,
-                data={
-                    **tool_call_log.data,
-                    "output": response_content,
-                    "files": len(tool_files),
-                    "meta": tool_invoke_meta.to_dict() if tool_invoke_meta else None,
-                },
-            )
-            final_content = response_content or "Tool executed successfully"
-            # Add tool response to messages
-            messages.append(
-                ToolPromptMessage(
-                    content=final_content,
-                    tool_call_id=tool_call_id,
-                    name=tool_name,
-                )
-            )
-            return response_content, tool_files, tool_invoke_meta
-        except Exception as e:
-            # Tool invocation failed, yield error log
-            error_message = str(e)
-            tool_call_log.status = AgentLog.LogStatus.ERROR
-            tool_call_log.error = error_message
-            tool_call_log.data = {
-                **tool_call_log.data,
-                "error": error_message,
-            }
-            yield tool_call_log
-
-            # Add error message to conversation
-            error_content = f"Tool execution failed: {error_message}"
-            messages.append(
-                ToolPromptMessage(
-                    content=error_content,
-                    tool_call_id=tool_call_id,
-                    name=tool_name,
-                )
-            )
-            return error_content, [], None
@@ -1,418 +0,0 @@
-"""ReAct strategy implementation."""
-
-from __future__ import annotations
-
-import json
-from collections.abc import Generator
-from typing import TYPE_CHECKING, Any, Union
-
-from core.agent.entities import AgentLog, AgentResult, AgentScratchpadUnit, ExecutionContext
-from core.agent.output_parser.cot_output_parser import CotAgentOutputParser
-from core.file import File
-from core.model_manager import ModelInstance
-from core.model_runtime.entities import (
-    AssistantPromptMessage,
-    LLMResult,
-    LLMResultChunk,
-    LLMResultChunkDelta,
-    PromptMessage,
-    SystemPromptMessage,
-)
-
-from .base import AgentPattern, ToolInvokeHook
-
-if TYPE_CHECKING:
-    from core.tools.__base.tool import Tool
-
-
-class ReActStrategy(AgentPattern):
-    """ReAct strategy using reasoning and acting approach."""
-
-    def __init__(
-        self,
-        model_instance: ModelInstance,
-        tools: list[Tool],
-        context: ExecutionContext,
-        max_iterations: int = 10,
-        workflow_call_depth: int = 0,
-        files: list[File] = [],
-        tool_invoke_hook: ToolInvokeHook | None = None,
-        instruction: str = "",
-    ):
-        """Initialize the ReAct strategy with instruction support."""
-        super().__init__(
-            model_instance=model_instance,
-            tools=tools,
-            context=context,
-            max_iterations=max_iterations,
-            workflow_call_depth=workflow_call_depth,
-            files=files,
-            tool_invoke_hook=tool_invoke_hook,
-        )
-        self.instruction = instruction
-
-    def run(
-        self,
-        prompt_messages: list[PromptMessage],
-        model_parameters: dict[str, Any],
-        stop: list[str] = [],
-        stream: bool = True,
-    ) -> Generator[LLMResultChunk | AgentLog, None, AgentResult]:
-        """Execute the ReAct agent strategy."""
-        # Initialize tracking
-        agent_scratchpad: list[AgentScratchpadUnit] = []
-        iteration_step: int = 1
-        max_iterations: int = self.max_iterations + 1
-        react_state: bool = True
-        total_usage: dict[str, Any] = {"usage": None}
-        output_files: list[File] = []  # Track files produced by tools
-        final_text: str = ""
-        finish_reason: str | None = None
-
-        # Add "Observation" to stop sequences
-        if "Observation" not in stop:
-            stop = stop.copy()
-            stop.append("Observation")
-
-        while react_state and iteration_step <= max_iterations:
-            react_state = False
-            round_log = self._create_log(
-                label=f"ROUND {iteration_step}",
-                log_type=AgentLog.LogType.ROUND,
-                status=AgentLog.LogStatus.START,
-                data={},
-            )
-            yield round_log
-
-            # Build prompt with/without tools based on iteration
-            include_tools = iteration_step < max_iterations
-            current_messages = self._build_prompt_with_react_format(
-                prompt_messages, agent_scratchpad, include_tools, self.instruction
-            )
-
-            model_log = self._create_log(
-                label=f"{self.model_instance.model} Thought",
-                log_type=AgentLog.LogType.THOUGHT,
-                status=AgentLog.LogStatus.START,
-                data={},
-                parent_id=round_log.id,
-                extra_metadata={
-                    AgentLog.LogMetadata.PROVIDER: self.model_instance.provider,
-                },
-            )
-            yield model_log
-
-            # Track usage for this round only
-            round_usage: dict[str, Any] = {"usage": None}
-
-            # Use current messages directly (files are handled by base class if needed)
-            messages_to_use = current_messages
-
-            # Invoke model
-            chunks: Union[Generator[LLMResultChunk, None, None], LLMResult] = self.model_instance.invoke_llm(
-                prompt_messages=messages_to_use,
-                model_parameters=model_parameters,
-                stop=stop,
-                stream=stream,
-                user=self.context.user_id or "",
-                callbacks=[],
-            )
-
-            # Process response
-            scratchpad, chunk_finish_reason = yield from self._handle_chunks(
-                chunks, round_usage, model_log, current_messages
-            )
-            agent_scratchpad.append(scratchpad)
-
-            # Accumulate to total usage
-            round_usage_value = round_usage.get("usage")
-            if round_usage_value:
-                self._accumulate_usage(total_usage, round_usage_value)
-
-            # Update finish reason
-            if chunk_finish_reason:
-                finish_reason = chunk_finish_reason
-
-            # Check if we have an action to execute
-            if scratchpad.action and scratchpad.action.action_name.lower() != "final answer":
-                react_state = True
-                # Execute tool
-                observation, tool_files = yield from self._handle_tool_call(
-                    scratchpad.action, current_messages, round_log
-                )
-                scratchpad.observation = observation
-                # Track files produced by tools
-                output_files.extend(tool_files)
-
-                # Add observation to scratchpad for display
-                yield self._create_text_chunk(f"\nObservation: {observation}\n", current_messages)
-            else:
-                # Extract final answer
-                if scratchpad.action and scratchpad.action.action_input:
-                    final_answer = scratchpad.action.action_input
-                    if isinstance(final_answer, dict):
-                        final_answer = json.dumps(final_answer, ensure_ascii=False)
-                    final_text = str(final_answer)
-                elif scratchpad.thought:
-                    # If no action but we have thought, use thought as final answer
-                    final_text = scratchpad.thought
-
-            yield self._finish_log(
-                round_log,
-                data={
-                    "thought": scratchpad.thought,
-                    "action": scratchpad.action_str if scratchpad.action else None,
-                    "observation": scratchpad.observation or None,
-                    "final_answer": final_text if not react_state else None,
-                },
-                usage=round_usage.get("usage"),
-            )
-            iteration_step += 1
-
-        # Return final result
-
-        from core.agent.entities import AgentResult
-
-        return AgentResult(
-            text=final_text, files=output_files, usage=total_usage.get("usage"), finish_reason=finish_reason
-        )
-
-    def _build_prompt_with_react_format(
-        self,
-        original_messages: list[PromptMessage],
-        agent_scratchpad: list[AgentScratchpadUnit],
-        include_tools: bool = True,
-        instruction: str = "",
-    ) -> list[PromptMessage]:
-        """Build prompt messages with ReAct format."""
-        # Copy messages to avoid modifying original
-        messages = list(original_messages)
-
-        # Find and update the system prompt that should already exist
-        system_prompt_found = False
-        for i, msg in enumerate(messages):
-            if isinstance(msg, SystemPromptMessage):
-                system_prompt_found = True
-                # The system prompt from frontend already has the template, just replace placeholders
-
-                # Format tools
-                tools_str = ""
-                tool_names = []
-                if include_tools and self.tools:
-                    # Convert tools to prompt message tools format
-                    prompt_tools = [tool.to_prompt_message_tool() for tool in self.tools]
-                    tool_names = [tool.name for tool in prompt_tools]
-
-                    # Format tools as JSON for comprehensive information
-                    from core.model_runtime.utils.encoders import jsonable_encoder
-
-                    tools_str = json.dumps(jsonable_encoder(prompt_tools), indent=2)
-                    tool_names_str = ", ".join(f'"{name}"' for name in tool_names)
-                else:
-                    tools_str = "No tools available"
-                    tool_names_str = ""
-
-                # Replace placeholders in the existing system prompt
-                updated_content = msg.content
-                assert isinstance(updated_content, str)
-                updated_content = updated_content.replace("{{instruction}}", instruction)
-                updated_content = updated_content.replace("{{tools}}", tools_str)
-                updated_content = updated_content.replace("{{tool_names}}", tool_names_str)
-
-                # Create new SystemPromptMessage with updated content
-                messages[i] = SystemPromptMessage(content=updated_content)
-                break
-
-        # If no system prompt found, that's unexpected but add scratchpad anyway
-        if not system_prompt_found:
-            # This shouldn't happen if frontend is working correctly
-            pass
-
-        # Format agent scratchpad
-        scratchpad_str = ""
-        if agent_scratchpad:
-            scratchpad_parts: list[str] = []
-            for unit in agent_scratchpad:
-                if unit.thought:
-                    scratchpad_parts.append(f"Thought: {unit.thought}")
-                if unit.action_str:
-                    scratchpad_parts.append(f"Action:\n```\n{unit.action_str}\n```")
-                if unit.observation:
-                    scratchpad_parts.append(f"Observation: {unit.observation}")
-            scratchpad_str = "\n".join(scratchpad_parts)
-
-        # If there's a scratchpad, append it to the last message
-        if scratchpad_str:
-            messages.append(AssistantPromptMessage(content=scratchpad_str))
-
-        return messages
-
-    def _handle_chunks(
-        self,
-        chunks: Union[Generator[LLMResultChunk, None, None], LLMResult],
-        llm_usage: dict[str, Any],
-        model_log: AgentLog,
-        current_messages: list[PromptMessage],
-    ) -> Generator[
-        LLMResultChunk | AgentLog,
-        None,
-        tuple[AgentScratchpadUnit, str | None],
-    ]:
-        """Handle LLM response chunks and extract action/thought.
-
-        Returns a tuple of (scratchpad_unit, finish_reason).
-        """
-        usage_dict: dict[str, Any] = {}
-
-        # Convert non-streaming to streaming format if needed
-        if isinstance(chunks, LLMResult):
-            # Create a generator from the LLMResult
-            def result_to_chunks() -> Generator[LLMResultChunk, None, None]:
-                yield LLMResultChunk(
-                    model=chunks.model,
-                    prompt_messages=chunks.prompt_messages,
-                    delta=LLMResultChunkDelta(
-                        index=0,
-                        message=chunks.message,
-                        usage=chunks.usage,
-                        finish_reason=None,  # LLMResult doesn't have finish_reason, only streaming chunks do
-                    ),
-                    system_fingerprint=chunks.system_fingerprint or "",
-                )
-
-            streaming_chunks = result_to_chunks()
-        else:
-            streaming_chunks = chunks
-
-        react_chunks = CotAgentOutputParser.handle_react_stream_output(streaming_chunks, usage_dict)
-
-        # Initialize scratchpad unit
-        scratchpad = AgentScratchpadUnit(
-            agent_response="",
-            thought="",
-            action_str="",
-            observation="",
-            action=None,
-        )
-
-        finish_reason: str | None = None
-
-        # Process chunks
-        for chunk in react_chunks:
-            if isinstance(chunk, AgentScratchpadUnit.Action):
-                # Action detected
-                action_str = json.dumps(chunk.model_dump())
-                scratchpad.agent_response = (scratchpad.agent_response or "") + action_str
-                scratchpad.action_str = action_str
-                scratchpad.action = chunk
-
-                yield self._create_text_chunk(json.dumps(chunk.model_dump()), current_messages)
-            else:
-                # Text chunk
-                chunk_text = str(chunk)
-                scratchpad.agent_response = (scratchpad.agent_response or "") + chunk_text
-                scratchpad.thought = (scratchpad.thought or "") + chunk_text
-
-                yield self._create_text_chunk(chunk_text, current_messages)
-
-        # Update usage
-        if usage_dict.get("usage"):
-            if llm_usage.get("usage"):
-                self._accumulate_usage(llm_usage, usage_dict["usage"])
-            else:
-                llm_usage["usage"] = usage_dict["usage"]
-
-        # Clean up thought
-        scratchpad.thought = (scratchpad.thought or "").strip() or "I am thinking about how to help you"
-
-        # Finish model log
-        yield self._finish_log(
-            model_log,
-            data={
-                "thought": scratchpad.thought,
-                "action": scratchpad.action_str if scratchpad.action else None,
-            },
-            usage=llm_usage.get("usage"),
-        )
-
-        return scratchpad, finish_reason
-
-    def _handle_tool_call(
-        self,
-        action: AgentScratchpadUnit.Action,
-        prompt_messages: list[PromptMessage],
-        round_log: AgentLog,
-    ) -> Generator[AgentLog, None, tuple[str, list[File]]]:
-        """Handle tool call and return observation with files."""
-        tool_name = action.action_name
-        tool_args: dict[str, Any] | str = action.action_input
-
-        # Find tool instance first to get metadata
-        tool_instance = self._find_tool_by_name(tool_name)
-        tool_metadata = self._get_tool_metadata(tool_instance) if tool_instance else {}
-
-        # Start tool log with tool metadata
-        tool_log = self._create_log(
-            label=f"CALL {tool_name}",
-            log_type=AgentLog.LogType.TOOL_CALL,
-            status=AgentLog.LogStatus.START,
-            data={
-                "tool_name": tool_name,
-                "tool_args": tool_args,
-            },
-            parent_id=round_log.id,
-            extra_metadata=tool_metadata,
-        )
-        yield tool_log
-
-        if not tool_instance:
-            # Finish tool log with error
-            yield self._finish_log(
-                tool_log,
-                data={
-                    **tool_log.data,
-                    "error": f"Tool {tool_name} not found",
-                },
-            )
-            return f"Tool {tool_name} not found", []
-
-        # Ensure tool_args is a dict
-        tool_args_dict: dict[str, Any]
-        if isinstance(tool_args, str):
-            try:
-                tool_args_dict = json.loads(tool_args)
-            except json.JSONDecodeError:
-                tool_args_dict = {"input": tool_args}
-        elif not isinstance(tool_args, dict):
-            tool_args_dict = {"input": str(tool_args)}
-        else:
-            tool_args_dict = tool_args
-
-        # Invoke tool using base class method with error handling
-        try:
-            response_content, tool_files, tool_invoke_meta = self._invoke_tool(tool_instance, tool_args_dict, tool_name)
-
-            # Finish tool log
-            yield self._finish_log(
-                tool_log,
-                data={
-                    **tool_log.data,
-                    "output": response_content,
-                    "files": len(tool_files),
-                    "meta": tool_invoke_meta.to_dict() if tool_invoke_meta else None,
-                },
-            )
-
-            return response_content or "Tool executed successfully", tool_files
-        except Exception as e:
-            # Tool invocation failed, yield error log
-            error_message = str(e)
-            tool_log.status = AgentLog.LogStatus.ERROR
-            tool_log.error = error_message
-            tool_log.data = {
-                **tool_log.data,
-                "error": error_message,
-            }
-            yield tool_log
-
-            return f"Tool execution failed: {error_message}", []
@@ -1,107 +0,0 @@
-"""Strategy factory for creating agent strategies."""
-
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-from core.agent.entities import AgentEntity, ExecutionContext
-from core.file.models import File
-from core.model_manager import ModelInstance
-from core.model_runtime.entities.model_entities import ModelFeature
-
-from .base import AgentPattern, ToolInvokeHook
-from .function_call import FunctionCallStrategy
-from .react import ReActStrategy
-
-if TYPE_CHECKING:
-    from core.tools.__base.tool import Tool
-
-
-class StrategyFactory:
-    """Factory for creating agent strategies based on model features."""
-
-    # Tool calling related features
-    TOOL_CALL_FEATURES = {ModelFeature.TOOL_CALL, ModelFeature.MULTI_TOOL_CALL, ModelFeature.STREAM_TOOL_CALL}
-
-    @staticmethod
-    def create_strategy(
-        model_features: list[ModelFeature],
-        model_instance: ModelInstance,
-        context: ExecutionContext,
-        tools: list[Tool],
-        files: list[File],
-        max_iterations: int = 10,
-        workflow_call_depth: int = 0,
-        agent_strategy: AgentEntity.Strategy | None = None,
-        tool_invoke_hook: ToolInvokeHook | None = None,
-        instruction: str = "",
-    ) -> AgentPattern:
-        """
-        Create an appropriate strategy based on model features.
-
-        Args:
-            model_features: List of model features/capabilities
-            model_instance: Model instance to use
-            context: Execution context containing trace/audit information
-            tools: Available tools
-            files: Available files
-            max_iterations: Maximum iterations for the strategy
-            workflow_call_depth: Depth of workflow calls
-            agent_strategy: Optional explicit strategy override
-            tool_invoke_hook: Optional hook for custom tool invocation (e.g., agent_invoke)
-            instruction: Optional instruction for ReAct strategy
-
-        Returns:
-            AgentStrategy instance
-        """
-        # If explicit strategy is provided and it's Function Calling, try to use it if supported
-        if agent_strategy == AgentEntity.Strategy.FUNCTION_CALLING:
-            if set(model_features) & StrategyFactory.TOOL_CALL_FEATURES:
-                return FunctionCallStrategy(
-                    model_instance=model_instance,
-                    context=context,
-                    tools=tools,
-                    files=files,
-                    max_iterations=max_iterations,
-                    workflow_call_depth=workflow_call_depth,
-                    tool_invoke_hook=tool_invoke_hook,
-                )
-            # Fallback to ReAct if FC is requested but not supported
-
-        # If explicit strategy is Chain of Thought (ReAct)
-        if agent_strategy == AgentEntity.Strategy.CHAIN_OF_THOUGHT:
-            return ReActStrategy(
-                model_instance=model_instance,
-                context=context,
-                tools=tools,
-                files=files,
-                max_iterations=max_iterations,
-                workflow_call_depth=workflow_call_depth,
-                tool_invoke_hook=tool_invoke_hook,
-                instruction=instruction,
-            )
-
-        # Default auto-selection logic
-        if set(model_features) & StrategyFactory.TOOL_CALL_FEATURES:
-            # Model supports native function calling
-            return FunctionCallStrategy(
-                model_instance=model_instance,
-                context=context,
-                tools=tools,
-                files=files,
-                max_iterations=max_iterations,
-                workflow_call_depth=workflow_call_depth,
-                tool_invoke_hook=tool_invoke_hook,
-            )
-        else:
-            # Use ReAct strategy for models without function calling
-            return ReActStrategy(
-                model_instance=model_instance,
-                context=context,
-                tools=tools,
-                files=files,
-                max_iterations=max_iterations,
-                workflow_call_depth=workflow_call_depth,
-                tool_invoke_hook=tool_invoke_hook,
-                instruction=instruction,
-            )
@@ -1,4 +1,3 @@
-import json
 from collections.abc import Sequence
 from enum import StrEnum, auto
 from typing import Any, Literal
@@ -121,7 +120,7 @@ class VariableEntity(BaseModel):
    allowed_file_types: Sequence[FileType] | None = Field(default_factory=list)
    allowed_file_extensions: Sequence[str] | None = Field(default_factory=list)
    allowed_file_upload_methods: Sequence[FileTransferMethod] | None = Field(default_factory=list)
-    json_schema: str | None = Field(default=None)
+    json_schema: dict | None = Field(default=None)

    @field_validator("description", mode="before")
    @classmethod
@@ -135,17 +134,11 @@ class VariableEntity(BaseModel):

    @field_validator("json_schema")
    @classmethod
-    def validate_json_schema(cls, schema: str | None) -> str | None:
+    def validate_json_schema(cls, schema: dict | None) -> dict | None:
        if schema is None:
            return None
-
        try:
-            json_schema = json.loads(schema)
-        except json.JSONDecodeError:
-            raise ValueError(f"invalid json_schema value {schema}")
-
-        try:
-            Draft7Validator.check_schema(json_schema)
+            Draft7Validator.check_schema(schema)
        except SchemaError as e:
            raise ValueError(f"Invalid JSON schema: {e.message}")
        return schema
@@ -26,7 +26,6 @@ class AdvancedChatAppConfigManager(BaseAppConfigManager):
    @classmethod
    def get_app_config(cls, app_model: App, workflow: Workflow) -> AdvancedChatAppConfig:
        features_dict = workflow.features_dict
-
        app_mode = AppMode.value_of(app_model.mode)
        app_config = AdvancedChatAppConfig(
            tenant_id=app_model.tenant_id,
@@ -24,7 +24,7 @@ from core.app.layers.conversation_variable_persist_layer import ConversationVari
 from core.db.session_factory import session_factory
 from core.moderation.base import ModerationError
 from core.moderation.input_moderation import InputModeration
-from core.variables.variables import VariableUnion
+from core.variables.variables import Variable
 from core.workflow.enums import WorkflowType
 from core.workflow.graph_engine.command_channels.redis_channel import RedisChannel
 from core.workflow.graph_engine.layers.base import GraphEngineLayer
@@ -39,7 +39,6 @@ from extensions.ext_database import db
 from extensions.ext_redis import redis_client
 from extensions.otel import WorkflowAppRunnerHandler, trace_span
 from models import Workflow
-from models.enums import UserFrom
 from models.model import App, Conversation, Message, MessageAnnotation
 from models.workflow import ConversationVariable
 from services.conversation_variable_updater import ConversationVariableUpdater
@@ -106,6 +105,11 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
        if not app_record:
            raise ValueError("App not found")

+        invoke_from = self.application_generate_entity.invoke_from
+        if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
+            invoke_from = InvokeFrom.DEBUGGER
+        user_from = self._resolve_user_from(invoke_from)
+
        if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
            # Handle single iteration or single loop run
            graph, variable_pool, graph_runtime_state = self._prepare_single_node_execution(
@@ -145,8 +149,8 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
                system_variables=system_inputs,
                user_inputs=inputs,
                environment_variables=self._workflow.environment_variables,
-                # Based on the definition of `VariableUnion`,
-                # `list[Variable]` can be safely used as `list[VariableUnion]` since they are compatible.
+                # Based on the definition of `Variable`,
+                # `VariableBase` instances can be safely used as `Variable` since they are compatible.
                conversation_variables=conversation_variables,
            )

@@ -158,6 +162,8 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
                workflow_id=self._workflow.id,
                tenant_id=self._workflow.tenant_id,
                user_id=self.application_generate_entity.user_id,
+                user_from=user_from,
+                invoke_from=invoke_from,
            )

        db.session.close()
@@ -175,12 +181,8 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
            graph=graph,
            graph_config=self._workflow.graph_dict,
            user_id=self.application_generate_entity.user_id,
-            user_from=(
-                UserFrom.ACCOUNT
-                if self.application_generate_entity.invoke_from in {InvokeFrom.EXPLORE, InvokeFrom.DEBUGGER}
-                else UserFrom.END_USER
-            ),
-            invoke_from=self.application_generate_entity.invoke_from,
+            user_from=user_from,
+            invoke_from=invoke_from,
            call_depth=self.application_generate_entity.call_depth,
            variable_pool=variable_pool,
            graph_runtime_state=graph_runtime_state,
@@ -316,7 +318,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
            trace_manager=app_generate_entity.trace_manager,
        )

-    def _initialize_conversation_variables(self) -> list[VariableUnion]:
+    def _initialize_conversation_variables(self) -> list[Variable]:
        """
        Initialize conversation variables for the current conversation.

@@ -341,7 +343,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
            conversation_variables = [var.to_variable() for var in existing_variables]

            session.commit()
-            return cast(list[VariableUnion], conversation_variables)
+            return cast(list[Variable], conversation_variables)

    def _load_existing_conversation_variables(self, session: Session) -> list[ConversationVariable]:
        """
@@ -82,7 +82,7 @@ class AdvancedChatAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk

    @classmethod
@@ -110,7 +110,7 @@ class AdvancedChatAppGenerateResponseConverter(AppGenerateResponseConverter):
            }

            if isinstance(sub_stream_response, MessageEndStreamResponse):
-                sub_stream_response_dict = sub_stream_response.model_dump(mode="json")
+                sub_stream_response_dict = sub_stream_response.model_dump(mode="json", exclude_none=True)
                metadata = sub_stream_response_dict.get("metadata", {})
                sub_stream_response_dict["metadata"] = cls._get_simple_metadata(metadata)
                response_chunk.update(sub_stream_response_dict)
@@ -120,6 +120,6 @@ class AdvancedChatAppGenerateResponseConverter(AppGenerateResponseConverter):
            elif isinstance(sub_stream_response, NodeStartStreamResponse | NodeFinishStreamResponse):
                response_chunk.update(sub_stream_response.to_ignore_detail_dict())
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))

            yield response_chunk
@@ -4,7 +4,6 @@ import re
 import time
 from collections.abc import Callable, Generator, Mapping
 from contextlib import contextmanager
-from dataclasses import dataclass, field
 from threading import Thread
 from typing import Any, Union

@@ -20,7 +19,6 @@ from core.app.entities.app_invoke_entities import (
    InvokeFrom,
 )
 from core.app.entities.queue_entities import (
-    ChunkType,
    MessageQueueMessage,
    QueueAdvancedChatMessageEndEvent,
    QueueAgentLogEvent,
@@ -72,122 +70,13 @@ from core.workflow.runtime import GraphRuntimeState
 from core.workflow.system_variable import SystemVariable
 from extensions.ext_database import db
 from libs.datetime_utils import naive_utc_now
-from models import Account, Conversation, EndUser, LLMGenerationDetail, Message, MessageFile
+from models import Account, Conversation, EndUser, Message, MessageFile
 from models.enums import CreatorUserRole
 from models.workflow import Workflow

 logger = logging.getLogger(__name__)


-@dataclass
-class StreamEventBuffer:
-    """
-    Buffer for recording stream events in order to reconstruct the generation sequence.
-    Records the exact order of text chunks, thoughts, and tool calls as they stream.
-    """
-
-    # Accumulated reasoning content (each thought block is a separate element)
-    reasoning_content: list[str] = field(default_factory=list)
-    # Current reasoning buffer (accumulates until we see a different event type)
-    _current_reasoning: str = ""
-    # Tool calls with their details
-    tool_calls: list[dict] = field(default_factory=list)
-    # Tool call ID to index mapping for updating results
-    _tool_call_id_map: dict[str, int] = field(default_factory=dict)
-    # Sequence of events in stream order
-    sequence: list[dict] = field(default_factory=list)
-    # Current position in answer text
-    _content_position: int = 0
-    # Track last event type to detect transitions
-    _last_event_type: str | None = None
-
-    def _flush_current_reasoning(self) -> None:
-        """Flush accumulated reasoning to the list and add to sequence."""
-        if self._current_reasoning.strip():
-            self.reasoning_content.append(self._current_reasoning.strip())
-            self.sequence.append({"type": "reasoning", "index": len(self.reasoning_content) - 1})
-            self._current_reasoning = ""
-
-    def record_text_chunk(self, text: str) -> None:
-        """Record a text chunk event."""
-        if not text:
-            return
-
-        # Flush any pending reasoning first
-        if self._last_event_type == "thought":
-            self._flush_current_reasoning()
-
-        text_len = len(text)
-        start_pos = self._content_position
-
-        # If last event was also content, extend it; otherwise create new
-        if self.sequence and self.sequence[-1].get("type") == "content":
-            self.sequence[-1]["end"] = start_pos + text_len
-        else:
-            self.sequence.append({"type": "content", "start": start_pos, "end": start_pos + text_len})
-
-        self._content_position += text_len
-        self._last_event_type = "content"
-
-    def record_thought_chunk(self, text: str) -> None:
-        """Record a thought/reasoning chunk event."""
-        if not text:
-            return
-
-        # Accumulate thought content
-        self._current_reasoning += text
-        self._last_event_type = "thought"
-
-    def record_tool_call(self, tool_call_id: str, tool_name: str, tool_arguments: str) -> None:
-        """Record a tool call event."""
-        if not tool_call_id:
-            return
-
-        # Flush any pending reasoning first
-        if self._last_event_type == "thought":
-            self._flush_current_reasoning()
-
-        # Check if this tool call already exists (we might get multiple chunks)
-        if tool_call_id in self._tool_call_id_map:
-            idx = self._tool_call_id_map[tool_call_id]
-            # Update arguments if provided
-            if tool_arguments:
-                self.tool_calls[idx]["arguments"] = tool_arguments
-        else:
-            # New tool call
-            tool_call = {
-                "id": tool_call_id or "",
-                "name": tool_name or "",
-                "arguments": tool_arguments or "",
-                "result": "",
-                "elapsed_time": None,
-            }
-            self.tool_calls.append(tool_call)
-            idx = len(self.tool_calls) - 1
-            self._tool_call_id_map[tool_call_id] = idx
-            self.sequence.append({"type": "tool_call", "index": idx})
-
-        self._last_event_type = "tool_call"
-
-    def record_tool_result(self, tool_call_id: str, result: str, tool_elapsed_time: float | None = None) -> None:
-        """Record a tool result event (update existing tool call)."""
-        if not tool_call_id:
-            return
-        if tool_call_id in self._tool_call_id_map:
-            idx = self._tool_call_id_map[tool_call_id]
-            self.tool_calls[idx]["result"] = result
-            self.tool_calls[idx]["elapsed_time"] = tool_elapsed_time
-
-    def finalize(self) -> None:
-        """Finalize the buffer, flushing any pending data."""
-        if self._last_event_type == "thought":
-            self._flush_current_reasoning()
-
-    def has_data(self) -> bool:
-        """Check if there's any meaningful data recorded."""
-        return bool(self.reasoning_content or self.tool_calls or self.sequence)
-
-
 class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
    """
    AdvancedChatAppGenerateTaskPipeline is a class that generate stream output and state management for Application.
@@ -255,8 +144,6 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        self._workflow_run_id: str = ""
        self._draft_var_saver_factory = draft_var_saver_factory
        self._graph_runtime_state: GraphRuntimeState | None = None
-        # Stream event buffer for recording generation sequence
-        self._stream_buffer = StreamEventBuffer()
        self._seed_graph_runtime_state_from_queue_manager()

    def process(self) -> Union[ChatbotAppBlockingResponse, Generator[ChatbotAppStreamResponse, None, None]]:
@@ -471,25 +358,6 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        if node_finish_resp:
            yield node_finish_resp

-        # For ANSWER nodes, check if we need to send a message_replace event
-        # Only send if the final output differs from the accumulated task_state.answer
-        # This happens when variables were updated by variable_assigner during workflow execution
-        if event.node_type == NodeType.ANSWER and event.outputs:
-            final_answer = event.outputs.get("answer")
-            if final_answer is not None and final_answer != self._task_state.answer:
-                logger.info(
-                    "ANSWER node final output '%s' differs from accumulated answer '%s', sending message_replace event",
-                    final_answer,
-                    self._task_state.answer,
-                )
-                # Update the task state answer
-                self._task_state.answer = str(final_answer)
-                # Send message_replace event to update the UI
-                yield self._message_cycle_manager.message_replace_to_stream_response(
-                    answer=str(final_answer),
-                    reason="variable_update",
-                )
-
    def _handle_node_failed_events(
        self,
        event: Union[QueueNodeFailedEvent, QueueNodeExceptionEvent],
@@ -515,7 +383,7 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        queue_message: Union[WorkflowQueueMessage, MessageQueueMessage] | None = None,
        **kwargs,
    ) -> Generator[StreamResponse, None, None]:
-        """Handle text chunk events and record to stream buffer for sequence reconstruction."""
+        """Handle text chunk events."""
        delta_text = event.text
        if delta_text is None:
            return
@@ -537,52 +405,9 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        if tts_publisher and queue_message:
            tts_publisher.publish(queue_message)

-        tool_call = event.tool_call
-        tool_result = event.tool_result
-        tool_payload = tool_call or tool_result
-        tool_call_id = tool_payload.id if tool_payload and tool_payload.id else ""
-        tool_name = tool_payload.name if tool_payload and tool_payload.name else ""
-        tool_arguments = tool_call.arguments if tool_call and tool_call.arguments else ""
-        tool_files = tool_result.files if tool_result else []
-        tool_elapsed_time = tool_result.elapsed_time if tool_result else None
-        tool_icon = tool_payload.icon if tool_payload else None
-        tool_icon_dark = tool_payload.icon_dark if tool_payload else None
-        # Record stream event based on chunk type
-        chunk_type = event.chunk_type or ChunkType.TEXT
-        match chunk_type:
-            case ChunkType.TEXT:
-                self._stream_buffer.record_text_chunk(delta_text)
-                self._task_state.answer += delta_text
-            case ChunkType.THOUGHT:
-                # Reasoning should not be part of final answer text
-                self._stream_buffer.record_thought_chunk(delta_text)
-            case ChunkType.TOOL_CALL:
-                self._stream_buffer.record_tool_call(
-                    tool_call_id=tool_call_id,
-                    tool_name=tool_name,
-                    tool_arguments=tool_arguments,
-                )
-            case ChunkType.TOOL_RESULT:
-                self._stream_buffer.record_tool_result(
-                    tool_call_id=tool_call_id,
-                    result=delta_text,
-                    tool_elapsed_time=tool_elapsed_time,
-                )
-                self._task_state.answer += delta_text
-            case _:
-                pass
+        self._task_state.answer += delta_text
        yield self._message_cycle_manager.message_to_stream_response(
-            answer=delta_text,
-            message_id=self._message_id,
-            from_variable_selector=event.from_variable_selector,
-            chunk_type=event.chunk_type.value if event.chunk_type else None,
-            tool_call_id=tool_call_id or None,
-            tool_name=tool_name or None,
-            tool_arguments=tool_arguments or None,
-            tool_files=tool_files,
-            tool_elapsed_time=tool_elapsed_time,
-            tool_icon=tool_icon,
-            tool_icon_dark=tool_icon_dark,
+            answer=delta_text, message_id=self._message_id, from_variable_selector=event.from_variable_selector
        )

    def _handle_iteration_start_event(
@@ -950,7 +775,6 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):

        # If there are assistant files, remove markdown image links from answer
        answer_text = self._task_state.answer
-        answer_text = self._strip_think_blocks(answer_text)
        if self._recorded_files:
            # Remove markdown image links since we're storing files separately
            answer_text = re.sub(r"!\[.*?\]\(.*?\)", "", answer_text).strip()
@@ -1002,54 +826,6 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        ]
        session.add_all(message_files)

-        # Save generation detail (reasoning/tool calls/sequence) from stream buffer
-        self._save_generation_detail(session=session, message=message)
-
-    @staticmethod
-    def _strip_think_blocks(text: str) -> str:
-        """Remove <think>...</think> blocks (including their content) from text."""
-        if not text or "<think" not in text.lower():
-            return text
-
-        clean_text = re.sub(r"<think[^>]*>.*?</think>", "", text, flags=re.IGNORECASE | re.DOTALL)
-        clean_text = re.sub(r"\n\s*\n", "\n\n", clean_text).strip()
-        return clean_text
-
-    def _save_generation_detail(self, *, session: Session, message: Message) -> None:
-        """
-        Save LLM generation detail for Chatflow using stream event buffer.
-        The buffer records the exact order of events as they streamed,
-        allowing accurate reconstruction of the generation sequence.
-        """
-        # Finalize the stream buffer to flush any pending data
-        self._stream_buffer.finalize()
-
-        # Only save if there's meaningful data
-        if not self._stream_buffer.has_data():
-            return
-
-        reasoning_content = self._stream_buffer.reasoning_content
-        tool_calls = self._stream_buffer.tool_calls
-        sequence = self._stream_buffer.sequence
-
-        # Check if generation detail already exists for this message
-        existing = session.query(LLMGenerationDetail).filter_by(message_id=message.id).first()
-
-        if existing:
-            existing.reasoning_content = json.dumps(reasoning_content) if reasoning_content else None
-            existing.tool_calls = json.dumps(tool_calls) if tool_calls else None
-            existing.sequence = json.dumps(sequence) if sequence else None
-        else:
-            generation_detail = LLMGenerationDetail(
-                tenant_id=self._application_generate_entity.app_config.tenant_id,
-                app_id=self._application_generate_entity.app_config.app_id,
-                message_id=message.id,
-                reasoning_content=json.dumps(reasoning_content) if reasoning_content else None,
-                tool_calls=json.dumps(tool_calls) if tool_calls else None,
-                sequence=json.dumps(sequence) if sequence else None,
-            )
-            session.add(generation_detail)
-
    def _seed_graph_runtime_state_from_queue_manager(self) -> None:
        """Bootstrap the cached runtime state from the queue manager when present."""
        candidate = self._base_task_pipeline.queue_manager.graph_runtime_state
@@ -3,8 +3,10 @@ from typing import cast

 from sqlalchemy import select

-from core.agent.agent_app_runner import AgentAppRunner
+from core.agent.cot_chat_agent_runner import CotChatAgentRunner
+from core.agent.cot_completion_agent_runner import CotCompletionAgentRunner
 from core.agent.entities import AgentEntity
+from core.agent.fc_agent_runner import FunctionCallAgentRunner
 from core.app.apps.agent_chat.app_config_manager import AgentChatAppConfig
 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.apps.base_app_runner import AppRunner
@@ -12,7 +14,8 @@ from core.app.entities.app_invoke_entities import AgentChatAppGenerateEntity
 from core.app.entities.queue_entities import QueueAnnotationReplyEvent
 from core.memory.token_buffer_memory import TokenBufferMemory
 from core.model_manager import ModelInstance
-from core.model_runtime.entities.model_entities import ModelFeature
+from core.model_runtime.entities.llm_entities import LLMMode
+from core.model_runtime.entities.model_entities import ModelFeature, ModelPropertyKey
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
 from core.moderation.base import ModerationError
 from extensions.ext_database import db
@@ -191,7 +194,22 @@ class AgentChatAppRunner(AppRunner):
            raise ValueError("Message not found")
        db.session.close()

-        runner = AgentAppRunner(
+        runner_cls: type[FunctionCallAgentRunner] | type[CotChatAgentRunner] | type[CotCompletionAgentRunner]
+        # start agent runner
+        if agent_entity.strategy == AgentEntity.Strategy.CHAIN_OF_THOUGHT:
+            # check LLM mode
+            if model_schema.model_properties.get(ModelPropertyKey.MODE) == LLMMode.CHAT:
+                runner_cls = CotChatAgentRunner
+            elif model_schema.model_properties.get(ModelPropertyKey.MODE) == LLMMode.COMPLETION:
+                runner_cls = CotCompletionAgentRunner
+            else:
+                raise ValueError(f"Invalid LLM mode: {model_schema.model_properties.get(ModelPropertyKey.MODE)}")
+        elif agent_entity.strategy == AgentEntity.Strategy.FUNCTION_CALLING:
+            runner_cls = FunctionCallAgentRunner
+        else:
+            raise ValueError(f"Invalid agent strategy: {agent_entity.strategy}")
+
+        runner = runner_cls(
            tenant_id=app_config.tenant_id,
            application_generate_entity=application_generate_entity,
            conversation=conversation_result,
@@ -81,7 +81,7 @@ class AgentChatAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk

    @classmethod
@@ -109,7 +109,7 @@ class AgentChatAppGenerateResponseConverter(AppGenerateResponseConverter):
            }

            if isinstance(sub_stream_response, MessageEndStreamResponse):
-                sub_stream_response_dict = sub_stream_response.model_dump(mode="json")
+                sub_stream_response_dict = sub_stream_response.model_dump(mode="json", exclude_none=True)
                metadata = sub_stream_response_dict.get("metadata", {})
                sub_stream_response_dict["metadata"] = cls._get_simple_metadata(metadata)
                response_chunk.update(sub_stream_response_dict)
@@ -117,6 +117,6 @@ class AgentChatAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))

            yield response_chunk
@@ -1,4 +1,3 @@
-import json
 from collections.abc import Generator, Mapping, Sequence
 from typing import TYPE_CHECKING, Any, Union, final

@@ -76,12 +75,24 @@ class BaseAppGenerator:
        user_inputs = {**user_inputs, **files_inputs, **file_list_inputs}

        # Check if all files are converted to File
-        if any(filter(lambda v: isinstance(v, dict), user_inputs.values())):
-            raise ValueError("Invalid input type")
-        if any(
-            filter(lambda v: isinstance(v, dict), filter(lambda item: isinstance(item, list), user_inputs.values()))
-        ):
-            raise ValueError("Invalid input type")
+        invalid_dict_keys = [
+            k
+            for k, v in user_inputs.items()
+            if isinstance(v, dict)
+            and entity_dictionary[k].type not in {VariableEntityType.FILE, VariableEntityType.JSON_OBJECT}
+        ]
+        if invalid_dict_keys:
+            raise ValueError(f"Invalid input type for {invalid_dict_keys}")
+
+        invalid_list_dict_keys = [
+            k
+            for k, v in user_inputs.items()
+            if isinstance(v, list)
+            and any(isinstance(item, dict) for item in v)
+            and entity_dictionary[k].type != VariableEntityType.FILE_LIST
+        ]
+        if invalid_list_dict_keys:
+            raise ValueError(f"Invalid input type for {invalid_list_dict_keys}")

        return user_inputs

@@ -178,12 +189,8 @@ class BaseAppGenerator:
                    elif value == 0:
                        value = False
            case VariableEntityType.JSON_OBJECT:
-                if not isinstance(value, str):
-                    raise ValueError(f"{variable_entity.variable} in input form must be a string")
-                try:
-                    json.loads(value)
-                except json.JSONDecodeError:
-                    raise ValueError(f"{variable_entity.variable} in input form must be a valid JSON object")
+                if value and not isinstance(value, dict):
+                    raise ValueError(f"{variable_entity.variable} in input form must be a dict")
            case _:
                raise AssertionError("this statement should be unreachable.")

@@ -81,7 +81,7 @@ class ChatAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk

    @classmethod
@@ -109,7 +109,7 @@ class ChatAppGenerateResponseConverter(AppGenerateResponseConverter):
            }

            if isinstance(sub_stream_response, MessageEndStreamResponse):
-                sub_stream_response_dict = sub_stream_response.model_dump(mode="json")
+                sub_stream_response_dict = sub_stream_response.model_dump(mode="json", exclude_none=True)
                metadata = sub_stream_response_dict.get("metadata", {})
                sub_stream_response_dict["metadata"] = cls._get_simple_metadata(metadata)
                response_chunk.update(sub_stream_response_dict)
@@ -117,6 +117,6 @@ class ChatAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))

            yield response_chunk
@@ -70,6 +70,8 @@ class _NodeSnapshot:
    """Empty string means the node is not executing inside an iteration."""
    loop_id: str = ""
    """Empty string means the node is not executing inside a loop."""
+    mention_parent_id: str = ""
+    """Empty string means the node is not an extractor node."""


 class WorkflowResponseConverter:
@@ -131,6 +133,7 @@ class WorkflowResponseConverter:
            start_at=event.start_at,
            iteration_id=event.in_iteration_id or "",
            loop_id=event.in_loop_id or "",
+            mention_parent_id=event.in_mention_parent_id or "",
        )
        node_execution_id = NodeExecutionId(event.node_execution_id)
        self._node_snapshots[node_execution_id] = snapshot
@@ -287,6 +290,7 @@ class WorkflowResponseConverter:
                created_at=int(snapshot.start_at.timestamp()),
                iteration_id=event.in_iteration_id,
                loop_id=event.in_loop_id,
+                mention_parent_id=event.in_mention_parent_id,
                agent_strategy=event.agent_strategy,
            ),
        )
@@ -373,6 +377,7 @@ class WorkflowResponseConverter:
                files=self.fetch_files_from_node_outputs(event.outputs or {}),
                iteration_id=event.in_iteration_id,
                loop_id=event.in_loop_id,
+                mention_parent_id=event.in_mention_parent_id,
            ),
        )

@@ -422,6 +427,7 @@ class WorkflowResponseConverter:
                files=self.fetch_files_from_node_outputs(event.outputs or {}),
                iteration_id=event.in_iteration_id,
                loop_id=event.in_loop_id,
+                mention_parent_id=event.in_mention_parent_id,
                retry_index=event.retry_index,
            ),
        )
@@ -671,7 +677,7 @@ class WorkflowResponseConverter:
            task_id=task_id,
            data=AgentLogStreamResponse.Data(
                node_execution_id=event.node_execution_id,
-                message_id=event.id,
+                id=event.id,
                parent_id=event.parent_id,
                label=event.label,
                error=event.error,
@@ -79,7 +79,7 @@ class CompletionAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk

    @classmethod
@@ -106,7 +106,7 @@ class CompletionAppGenerateResponseConverter(AppGenerateResponseConverter):
            }

            if isinstance(sub_stream_response, MessageEndStreamResponse):
-                sub_stream_response_dict = sub_stream_response.model_dump(mode="json")
+                sub_stream_response_dict = sub_stream_response.model_dump(mode="json", exclude_none=True)
                metadata = sub_stream_response_dict.get("metadata", {})
                if not isinstance(metadata, dict):
                    metadata = {}
@@ -116,6 +116,6 @@ class CompletionAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))

            yield response_chunk
@@ -60,7 +60,7 @@ class WorkflowAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(cast(dict, data))
            else:
-                response_chunk.update(sub_stream_response.model_dump())
+                response_chunk.update(sub_stream_response.model_dump(exclude_none=True))
            yield response_chunk

    @classmethod
@@ -91,5 +91,5 @@ class WorkflowAppGenerateResponseConverter(AppGenerateResponseConverter):
            elif isinstance(sub_stream_response, NodeStartStreamResponse | NodeFinishStreamResponse):
                response_chunk.update(cast(dict, sub_stream_response.to_ignore_detail_dict()))
            else:
-                response_chunk.update(sub_stream_response.model_dump())
+                response_chunk.update(sub_stream_response.model_dump(exclude_none=True))
            yield response_chunk
@@ -73,9 +73,15 @@ class PipelineRunner(WorkflowBasedAppRunner):
        """
        app_config = self.application_generate_entity.app_config
        app_config = cast(PipelineConfig, app_config)
+        invoke_from = self.application_generate_entity.invoke_from
+
+        if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
+            invoke_from = InvokeFrom.DEBUGGER
+
+        user_from = self._resolve_user_from(invoke_from)

        user_id = None
-        if self.application_generate_entity.invoke_from in {InvokeFrom.WEB_APP, InvokeFrom.SERVICE_API}:
+        if invoke_from in {InvokeFrom.WEB_APP, InvokeFrom.SERVICE_API}:
            end_user = db.session.query(EndUser).where(EndUser.id == self.application_generate_entity.user_id).first()
            if end_user:
                user_id = end_user.session_id
@@ -117,7 +123,7 @@ class PipelineRunner(WorkflowBasedAppRunner):
                dataset_id=self.application_generate_entity.dataset_id,
                datasource_type=self.application_generate_entity.datasource_type,
                datasource_info=self.application_generate_entity.datasource_info,
-                invoke_from=self.application_generate_entity.invoke_from.value,
+                invoke_from=invoke_from.value,
            )

            rag_pipeline_variables = []
@@ -149,6 +155,8 @@ class PipelineRunner(WorkflowBasedAppRunner):
                graph_runtime_state=graph_runtime_state,
                start_node_id=self.application_generate_entity.start_node_id,
                workflow=workflow,
+                user_from=user_from,
+                invoke_from=invoke_from,
            )

        # RUN WORKFLOW
@@ -159,12 +167,8 @@ class PipelineRunner(WorkflowBasedAppRunner):
            graph=graph,
            graph_config=workflow.graph_dict,
            user_id=self.application_generate_entity.user_id,
-            user_from=(
-                UserFrom.ACCOUNT
-                if self.application_generate_entity.invoke_from in {InvokeFrom.EXPLORE, InvokeFrom.DEBUGGER}
-                else UserFrom.END_USER
-            ),
-            invoke_from=self.application_generate_entity.invoke_from,
+            user_from=user_from,
+            invoke_from=invoke_from,
            call_depth=self.application_generate_entity.call_depth,
            graph_runtime_state=graph_runtime_state,
            variable_pool=variable_pool,
@@ -210,7 +214,12 @@ class PipelineRunner(WorkflowBasedAppRunner):
        return workflow

    def _init_rag_pipeline_graph(
-        self, workflow: Workflow, graph_runtime_state: GraphRuntimeState, start_node_id: str | None = None
+        self,
+        workflow: Workflow,
+        graph_runtime_state: GraphRuntimeState,
+        start_node_id: str | None = None,
+        user_from: UserFrom = UserFrom.ACCOUNT,
+        invoke_from: InvokeFrom = InvokeFrom.SERVICE_API,
    ) -> Graph:
        """
        Init pipeline graph
@@ -253,8 +262,8 @@ class PipelineRunner(WorkflowBasedAppRunner):
            workflow_id=workflow.id,
            graph_config=graph_config,
            user_id=self.application_generate_entity.user_id,
-            user_from=UserFrom.ACCOUNT,
-            invoke_from=InvokeFrom.SERVICE_API,
+            user_from=user_from,
+            invoke_from=invoke_from,
            call_depth=0,
        )

@@ -20,7 +20,6 @@ from core.workflow.workflow_entry import WorkflowEntry
 from extensions.ext_redis import redis_client
 from extensions.otel import WorkflowAppRunnerHandler, trace_span
 from libs.datetime_utils import naive_utc_now
-from models.enums import UserFrom
 from models.workflow import Workflow

 logger = logging.getLogger(__name__)
@@ -74,7 +73,12 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
            workflow_execution_id=self.application_generate_entity.workflow_execution_id,
        )

+        invoke_from = self.application_generate_entity.invoke_from
        # if only single iteration or single loop run is requested
+        if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
+            invoke_from = InvokeFrom.DEBUGGER
+        user_from = self._resolve_user_from(invoke_from)
+
        if self.application_generate_entity.single_iteration_run or self.application_generate_entity.single_loop_run:
            graph, variable_pool, graph_runtime_state = self._prepare_single_node_execution(
                workflow=self._workflow,
@@ -102,6 +106,8 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
                workflow_id=self._workflow.id,
                tenant_id=self._workflow.tenant_id,
                user_id=self.application_generate_entity.user_id,
+                user_from=user_from,
+                invoke_from=invoke_from,
                root_node_id=self._root_node_id,
            )

@@ -120,12 +126,8 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
            graph=graph,
            graph_config=self._workflow.graph_dict,
            user_id=self.application_generate_entity.user_id,
-            user_from=(
-                UserFrom.ACCOUNT
-                if self.application_generate_entity.invoke_from in {InvokeFrom.EXPLORE, InvokeFrom.DEBUGGER}
-                else UserFrom.END_USER
-            ),
-            invoke_from=self.application_generate_entity.invoke_from,
+            user_from=user_from,
+            invoke_from=invoke_from,
            call_depth=self.application_generate_entity.call_depth,
            variable_pool=variable_pool,
            graph_runtime_state=graph_runtime_state,
@@ -60,7 +60,7 @@ class WorkflowAppGenerateResponseConverter(AppGenerateResponseConverter):
                data = cls._error_to_stream_response(sub_stream_response.err)
                response_chunk.update(data)
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk

    @classmethod
@@ -91,5 +91,5 @@ class WorkflowAppGenerateResponseConverter(AppGenerateResponseConverter):
            elif isinstance(sub_stream_response, NodeStartStreamResponse | NodeFinishStreamResponse):
                response_chunk.update(sub_stream_response.to_ignore_detail_dict())
            else:
-                response_chunk.update(sub_stream_response.model_dump(mode="json"))
+                response_chunk.update(sub_stream_response.model_dump(mode="json", exclude_none=True))
            yield response_chunk
@@ -13,7 +13,6 @@ from core.app.apps.common.workflow_response_converter import WorkflowResponseCon
 from core.app.entities.app_invoke_entities import InvokeFrom, WorkflowAppGenerateEntity
 from core.app.entities.queue_entities import (
    AppQueueEvent,
-    ChunkType,
    MessageQueueMessage,
    QueueAgentLogEvent,
    QueueErrorEvent,
@@ -484,33 +483,11 @@ class WorkflowAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        if delta_text is None:
            return

-        tool_call = event.tool_call
-        tool_result = event.tool_result
-        tool_payload = tool_call or tool_result
-        tool_call_id = tool_payload.id if tool_payload and tool_payload.id else None
-        tool_name = tool_payload.name if tool_payload and tool_payload.name else None
-        tool_arguments = tool_call.arguments if tool_call else None
-        tool_elapsed_time = tool_result.elapsed_time if tool_result else None
-        tool_files = tool_result.files if tool_result else []
-        tool_icon = tool_payload.icon if tool_payload else None
-        tool_icon_dark = tool_payload.icon_dark if tool_payload else None
-
        # only publish tts message at text chunk streaming
        if tts_publisher and queue_message:
            tts_publisher.publish(queue_message)

-        yield self._text_chunk_to_stream_response(
-            text=delta_text,
-            from_variable_selector=event.from_variable_selector,
-            chunk_type=event.chunk_type,
-            tool_call_id=tool_call_id,
-            tool_name=tool_name,
-            tool_arguments=tool_arguments,
-            tool_files=tool_files,
-            tool_elapsed_time=tool_elapsed_time,
-            tool_icon=tool_icon,
-            tool_icon_dark=tool_icon_dark,
-        )
+        yield self._text_chunk_to_stream_response(delta_text, from_variable_selector=event.from_variable_selector)

    def _handle_agent_log_event(self, event: QueueAgentLogEvent, **kwargs) -> Generator[StreamResponse, None, None]:
        """Handle agent log events."""
@@ -673,61 +650,16 @@ class WorkflowAppGenerateTaskPipeline(GraphRuntimeStateSupport):
        session.add(workflow_app_log)

    def _text_chunk_to_stream_response(
-        self,
-        text: str,
-        from_variable_selector: list[str] | None = None,
-        chunk_type: ChunkType | None = None,
-        tool_call_id: str | None = None,
-        tool_name: str | None = None,
-        tool_arguments: str | None = None,
-        tool_files: list[str] | None = None,
-        tool_error: str | None = None,
-        tool_elapsed_time: float | None = None,
-        tool_icon: str | dict | None = None,
-        tool_icon_dark: str | dict | None = None,
+        self, text: str, from_variable_selector: list[str] | None = None
    ) -> TextChunkStreamResponse:
        """
        Handle completed event.
        :param text: text
        :return:
        """
-        from core.app.entities.task_entities import ChunkType as ResponseChunkType
-
-        response_chunk_type = ResponseChunkType(chunk_type.value) if chunk_type else ResponseChunkType.TEXT
-
-        data = TextChunkStreamResponse.Data(
-            text=text,
-            from_variable_selector=from_variable_selector,
-            chunk_type=response_chunk_type,
-        )
-
-        if response_chunk_type == ResponseChunkType.TOOL_CALL:
-            data = data.model_copy(
-                update={
-                    "tool_call_id": tool_call_id,
-                    "tool_name": tool_name,
-                    "tool_arguments": tool_arguments,
-                    "tool_icon": tool_icon,
-                    "tool_icon_dark": tool_icon_dark,
-                }
-            )
-        elif response_chunk_type == ResponseChunkType.TOOL_RESULT:
-            data = data.model_copy(
-                update={
-                    "tool_call_id": tool_call_id,
-                    "tool_name": tool_name,
-                    "tool_arguments": tool_arguments,
-                    "tool_files": tool_files,
-                    "tool_error": tool_error,
-                    "tool_elapsed_time": tool_elapsed_time,
-                    "tool_icon": tool_icon,
-                    "tool_icon_dark": tool_icon_dark,
-                }
-            )
-
        response = TextChunkStreamResponse(
            task_id=self._application_generate_entity.task_id,
-            data=data,
+            data=TextChunkStreamResponse.Data(text=text, from_variable_selector=from_variable_selector),
        )

        return response
@@ -77,10 +77,18 @@ class WorkflowBasedAppRunner:
        self._app_id = app_id
        self._graph_engine_layers = graph_engine_layers

+    @staticmethod
+    def _resolve_user_from(invoke_from: InvokeFrom) -> UserFrom:
+        if invoke_from in {InvokeFrom.EXPLORE, InvokeFrom.DEBUGGER}:
+            return UserFrom.ACCOUNT
+        return UserFrom.END_USER
+
    def _init_graph(
        self,
        graph_config: Mapping[str, Any],
        graph_runtime_state: GraphRuntimeState,
+        user_from: UserFrom,
+        invoke_from: InvokeFrom,
        workflow_id: str = "",
        tenant_id: str = "",
        user_id: str = "",
@@ -105,8 +113,8 @@ class WorkflowBasedAppRunner:
            workflow_id=workflow_id,
            graph_config=graph_config,
            user_id=user_id,
-            user_from=UserFrom.ACCOUNT,
-            invoke_from=InvokeFrom.SERVICE_API,
+            user_from=user_from,
+            invoke_from=invoke_from,
            call_depth=0,
        )

@@ -250,7 +258,7 @@ class WorkflowBasedAppRunner:
            graph_config=graph_config,
            user_id="",
            user_from=UserFrom.ACCOUNT,
-            invoke_from=InvokeFrom.SERVICE_API,
+            invoke_from=InvokeFrom.DEBUGGER,
            call_depth=0,
        )

@@ -377,6 +385,7 @@ class WorkflowBasedAppRunner:
                    start_at=event.start_at,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                    inputs=inputs,
                    process_data=process_data,
                    outputs=outputs,
@@ -397,6 +406,7 @@ class WorkflowBasedAppRunner:
                    start_at=event.start_at,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                    agent_strategy=event.agent_strategy,
                    provider_type=event.provider_type,
                    provider_id=event.provider_id,
@@ -420,6 +430,7 @@ class WorkflowBasedAppRunner:
                    execution_metadata=execution_metadata,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                )
            )
        elif isinstance(event, NodeRunFailedEvent):
@@ -436,6 +447,7 @@ class WorkflowBasedAppRunner:
                    execution_metadata=event.node_run_result.metadata,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                )
            )
        elif isinstance(event, NodeRunExceptionEvent):
@@ -452,23 +464,17 @@ class WorkflowBasedAppRunner:
                    execution_metadata=event.node_run_result.metadata,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                )
            )
        elif isinstance(event, NodeRunStreamChunkEvent):
-            from core.app.entities.queue_entities import ChunkType as QueueChunkType
-
-            if event.is_final and not event.chunk:
-                return
-
            self._publish_event(
                QueueTextChunkEvent(
                    text=event.chunk,
                    from_variable_selector=list(event.selector),
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
-                    chunk_type=QueueChunkType(event.chunk_type.value),
-                    tool_call=event.tool_call,
-                    tool_result=event.tool_result,
+                    in_mention_parent_id=event.in_mention_parent_id,
                )
            )
        elif isinstance(event, NodeRunRetrieverResourceEvent):
@@ -477,6 +483,7 @@ class WorkflowBasedAppRunner:
                    retriever_resources=event.retriever_resources,
                    in_iteration_id=event.in_iteration_id,
                    in_loop_id=event.in_loop_id,
+                    in_mention_parent_id=event.in_mention_parent_id,
                )
            )
        elif isinstance(event, NodeRunAgentLogEvent):
@@ -1,70 +0,0 @@
-"""
-LLM Generation Detail entities.
-
-Defines the structure for storing and transmitting LLM generation details
-including reasoning content, tool calls, and their sequence.
-"""
-
-from typing import Literal
-
-from pydantic import BaseModel, Field
-
-
-class ContentSegment(BaseModel):
-    """Represents a content segment in the generation sequence."""
-
-    type: Literal["content"] = "content"
-    start: int = Field(..., description="Start position in the text")
-    end: int = Field(..., description="End position in the text")
-
-
-class ReasoningSegment(BaseModel):
-    """Represents a reasoning segment in the generation sequence."""
-
-    type: Literal["reasoning"] = "reasoning"
-    index: int = Field(..., description="Index into reasoning_content array")
-
-
-class ToolCallSegment(BaseModel):
-    """Represents a tool call segment in the generation sequence."""
-
-    type: Literal["tool_call"] = "tool_call"
-    index: int = Field(..., description="Index into tool_calls array")
-
-
-SequenceSegment = ContentSegment | ReasoningSegment | ToolCallSegment
-
-
-class ToolCallDetail(BaseModel):
-    """Represents a tool call with its arguments and result."""
-
-    id: str = Field(default="", description="Unique identifier for the tool call")
-    name: str = Field(..., description="Name of the tool")
-    arguments: str = Field(default="", description="JSON string of tool arguments")
-    result: str = Field(default="", description="Result from the tool execution")
-    elapsed_time: float | None = Field(default=None, description="Elapsed time in seconds")
-
-
-class LLMGenerationDetailData(BaseModel):
-    """
-    Domain model for LLM generation detail.
-
-    Contains the structured data for reasoning content, tool calls,
-    and their display sequence.
-    """
-
-    reasoning_content: list[str] = Field(default_factory=list, description="List of reasoning segments")
-    tool_calls: list[ToolCallDetail] = Field(default_factory=list, description="List of tool call details")
-    sequence: list[SequenceSegment] = Field(default_factory=list, description="Display order of segments")
-
-    def is_empty(self) -> bool:
-        """Check if there's any meaningful generation detail."""
-        return not self.reasoning_content and not self.tool_calls
-
-    def to_response_dict(self) -> dict:
-        """Convert to dictionary for API response."""
-        return {
-            "reasoning_content": self.reasoning_content,
-            "tool_calls": [tc.model_dump() for tc in self.tool_calls],
-            "sequence": [seg.model_dump() for seg in self.sequence],
-        }
@@ -7,7 +7,7 @@ from pydantic import BaseModel, ConfigDict, Field

 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk
 from core.rag.entities.citation_metadata import RetrievalSourceMetadata
-from core.workflow.entities import AgentNodeStrategyInit, ToolCall, ToolResult
+from core.workflow.entities import AgentNodeStrategyInit
 from core.workflow.enums import WorkflowNodeExecutionMetadataKey
 from core.workflow.nodes import NodeType

@@ -177,17 +177,6 @@ class QueueLoopCompletedEvent(AppQueueEvent):
    error: str | None = None


-class ChunkType(StrEnum):
-    """Stream chunk type for LLM-related events."""
-
-    TEXT = "text"  # Normal text streaming
-    TOOL_CALL = "tool_call"  # Tool call arguments streaming
-    TOOL_RESULT = "tool_result"  # Tool execution result
-    THOUGHT = "thought"  # Agent thinking process (ReAct)
-    THOUGHT_START = "thought_start"  # Agent thought start
-    THOUGHT_END = "thought_end"  # Agent thought end
-
-
 class QueueTextChunkEvent(AppQueueEvent):
    """
    QueueTextChunkEvent entity
@@ -201,16 +190,8 @@ class QueueTextChunkEvent(AppQueueEvent):
    """iteration id if node is in iteration"""
    in_loop_id: str | None = None
    """loop id if node is in loop"""
-
-    # Extended fields for Agent/Tool streaming
-    chunk_type: ChunkType = ChunkType.TEXT
-    """type of the chunk"""
-
-    # Tool streaming payloads
-    tool_call: ToolCall | None = None
-    """structured tool call info"""
-    tool_result: ToolResult | None = None
-    """structured tool result info"""
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""


 class QueueAgentMessageEvent(AppQueueEvent):
@@ -250,6 +231,8 @@ class QueueRetrieverResourcesEvent(AppQueueEvent):
    """iteration id if node is in iteration"""
    in_loop_id: str | None = None
    """loop id if node is in loop"""
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""


 class QueueAnnotationReplyEvent(AppQueueEvent):
@@ -327,6 +310,8 @@ class QueueNodeStartedEvent(AppQueueEvent):
    node_run_index: int = 1  # FIXME(-LAN-): may not used
    in_iteration_id: str | None = None
    in_loop_id: str | None = None
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""
    start_at: datetime
    agent_strategy: AgentNodeStrategyInit | None = None

@@ -349,6 +334,8 @@ class QueueNodeSucceededEvent(AppQueueEvent):
    """iteration id if node is in iteration"""
    in_loop_id: str | None = None
    """loop id if node is in loop"""
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""
    start_at: datetime

    inputs: Mapping[str, object] = Field(default_factory=dict)
@@ -404,6 +391,8 @@ class QueueNodeExceptionEvent(AppQueueEvent):
    """iteration id if node is in iteration"""
    in_loop_id: str | None = None
    """loop id if node is in loop"""
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""
    start_at: datetime

    inputs: Mapping[str, object] = Field(default_factory=dict)
@@ -428,6 +417,8 @@ class QueueNodeFailedEvent(AppQueueEvent):
    """iteration id if node is in iteration"""
    in_loop_id: str | None = None
    """loop id if node is in loop"""
+    in_mention_parent_id: str | None = None
+    """parent node id if this is an extractor node event"""
    start_at: datetime

    inputs: Mapping[str, object] = Field(default_factory=dict)
@@ -113,38 +113,6 @@ class MessageStreamResponse(StreamResponse):
    answer: str
    from_variable_selector: list[str] | None = None

-    # Extended fields for Agent/Tool streaming (imported at runtime to avoid circular import)
-    chunk_type: str | None = None
-    """type of the chunk: text, tool_call, tool_result, thought"""
-
-    # Tool call fields (when chunk_type == "tool_call")
-    tool_call_id: str | None = None
-    """unique identifier for this tool call"""
-    tool_name: str | None = None
-    """name of the tool being called"""
-    tool_arguments: str | None = None
-    """accumulated tool arguments JSON"""
-
-    # Tool result fields (when chunk_type == "tool_result")
-    tool_files: list[str] | None = None
-    """file IDs produced by tool"""
-    tool_error: str | None = None
-    """error message if tool failed"""
-    tool_elapsed_time: float | None = None
-    """elapsed time spent executing the tool"""
-    tool_icon: str | dict | None = None
-    """icon of the tool"""
-    tool_icon_dark: str | dict | None = None
-    """dark theme icon of the tool"""
-
-    def model_dump(self, *args, **kwargs) -> dict[str, object]:
-        kwargs.setdefault("exclude_none", True)
-        return super().model_dump(*args, **kwargs)
-
-    def model_dump_json(self, *args, **kwargs) -> str:
-        kwargs.setdefault("exclude_none", True)
-        return super().model_dump_json(*args, **kwargs)
-

 class MessageAudioStreamResponse(StreamResponse):
    """
@@ -294,6 +262,7 @@ class NodeStartStreamResponse(StreamResponse):
        extras: dict[str, object] = Field(default_factory=dict)
        iteration_id: str | None = None
        loop_id: str | None = None
+        mention_parent_id: str | None = None
        agent_strategy: AgentNodeStrategyInit | None = None

    event: StreamEvent = StreamEvent.NODE_STARTED
@@ -317,6 +286,7 @@ class NodeStartStreamResponse(StreamResponse):
                "extras": {},
                "iteration_id": self.data.iteration_id,
                "loop_id": self.data.loop_id,
+                "mention_parent_id": self.data.mention_parent_id,
            },
        }

@@ -352,6 +322,7 @@ class NodeFinishStreamResponse(StreamResponse):
        files: Sequence[Mapping[str, Any]] | None = []
        iteration_id: str | None = None
        loop_id: str | None = None
+        mention_parent_id: str | None = None

    event: StreamEvent = StreamEvent.NODE_FINISHED
    workflow_run_id: str
@@ -381,6 +352,7 @@ class NodeFinishStreamResponse(StreamResponse):
                "files": [],
                "iteration_id": self.data.iteration_id,
                "loop_id": self.data.loop_id,
+                "mention_parent_id": self.data.mention_parent_id,
            },
        }

@@ -416,6 +388,7 @@ class NodeRetryStreamResponse(StreamResponse):
        files: Sequence[Mapping[str, Any]] | None = []
        iteration_id: str | None = None
        loop_id: str | None = None
+        mention_parent_id: str | None = None
        retry_index: int = 0

    event: StreamEvent = StreamEvent.NODE_RETRY
@@ -446,6 +419,7 @@ class NodeRetryStreamResponse(StreamResponse):
                "files": [],
                "iteration_id": self.data.iteration_id,
                "loop_id": self.data.loop_id,
+                "mention_parent_id": self.data.mention_parent_id,
                "retry_index": self.data.retry_index,
            },
        }
@@ -614,17 +588,6 @@ class LoopNodeCompletedStreamResponse(StreamResponse):
    data: Data


-class ChunkType(StrEnum):
-    """Stream chunk type for LLM-related events."""
-
-    TEXT = "text"  # Normal text streaming
-    TOOL_CALL = "tool_call"  # Tool call arguments streaming
-    TOOL_RESULT = "tool_result"  # Tool execution result
-    THOUGHT = "thought"  # Agent thinking process (ReAct)
-    THOUGHT_START = "thought_start"  # Agent thought start
-    THOUGHT_END = "thought_end"  # Agent thought end
-
-
 class TextChunkStreamResponse(StreamResponse):
    """
    TextChunkStreamResponse entity
@@ -638,36 +601,6 @@ class TextChunkStreamResponse(StreamResponse):
        text: str
        from_variable_selector: list[str] | None = None

-        # Extended fields for Agent/Tool streaming
-        chunk_type: ChunkType = ChunkType.TEXT
-        """type of the chunk"""
-
-        # Tool call fields (when chunk_type == TOOL_CALL)
-        tool_call_id: str | None = None
-        """unique identifier for this tool call"""
-        tool_name: str | None = None
-        """name of the tool being called"""
-        tool_arguments: str | None = None
-        """accumulated tool arguments JSON"""
-
-        # Tool result fields (when chunk_type == TOOL_RESULT)
-        tool_files: list[str] | None = None
-        """file IDs produced by tool"""
-        tool_error: str | None = None
-        """error message if tool failed"""
-
-        # Tool elapsed time fields (when chunk_type == TOOL_RESULT)
-        tool_elapsed_time: float | None = None
-        """elapsed time spent executing the tool"""
-
-        def model_dump(self, *args, **kwargs) -> dict[str, object]:
-            kwargs.setdefault("exclude_none", True)
-            return super().model_dump(*args, **kwargs)
-
-        def model_dump_json(self, *args, **kwargs) -> str:
-            kwargs.setdefault("exclude_none", True)
-            return super().model_dump_json(*args, **kwargs)
-
    event: StreamEvent = StreamEvent.TEXT_CHUNK
    data: Data

@@ -816,7 +749,7 @@ class AgentLogStreamResponse(StreamResponse):
        """

        node_execution_id: str
-        message_id: str
+        id: str
        label: str
        parent_id: str | None = None
        error: str | None = None
@@ -1,6 +1,6 @@
 import logging

-from core.variables import Variable
+from core.variables import VariableBase
 from core.workflow.constants import CONVERSATION_VARIABLE_NODE_ID
 from core.workflow.conversation_variable_updater import ConversationVariableUpdater
 from core.workflow.enums import NodeType
@@ -44,7 +44,7 @@ class ConversationVariablePersistenceLayer(GraphEngineLayer):
            if selector[0] != CONVERSATION_VARIABLE_NODE_ID:
                continue
            variable = self.graph_runtime_state.variable_pool.get(selector)
-            if not isinstance(variable, Variable):
+            if not isinstance(variable, VariableBase):
                logger.warning(
                    "Conversation variable not found in variable pool. selector=%s",
                    selector,
@@ -1,5 +1,4 @@
 import logging
-import re
 import time
 from collections.abc import Generator
 from threading import Thread
@@ -59,7 +58,7 @@ from core.prompt.utils.prompt_template_parser import PromptTemplateParser
 from events.message_event import message_was_created
 from extensions.ext_database import db
 from libs.datetime_utils import naive_utc_now
-from models.model import AppMode, Conversation, LLMGenerationDetail, Message, MessageAgentThought
+from models.model import AppMode, Conversation, Message, MessageAgentThought

 logger = logging.getLogger(__name__)

@@ -69,8 +68,6 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline):
    EasyUIBasedGenerateTaskPipeline is a class that generate stream output and state management for Application.
    """

-    _THINK_PATTERN = re.compile(r"<think[^>]*>(.*?)</think>", re.IGNORECASE | re.DOTALL)
-
    _task_state: EasyUITaskState
    _application_generate_entity: Union[ChatAppGenerateEntity, CompletionAppGenerateEntity, AgentChatAppGenerateEntity]

@@ -412,136 +409,11 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline):
                )
            )

-        # Save LLM generation detail if there's reasoning_content
-        self._save_generation_detail(session=session, message=message, llm_result=llm_result)
-
        message_was_created.send(
            message,
            application_generate_entity=self._application_generate_entity,
        )

-    def _save_generation_detail(self, *, session: Session, message: Message, llm_result: LLMResult) -> None:
-        """
-        Save LLM generation detail for Completion/Chat/Agent-Chat applications.
-        For Agent-Chat, also merges MessageAgentThought records.
-        """
-        import json
-
-        reasoning_list: list[str] = []
-        tool_calls_list: list[dict] = []
-        sequence: list[dict] = []
-        answer = message.answer or ""
-
-        # Check if this is Agent-Chat mode by looking for agent thoughts
-        agent_thoughts = (
-            session.query(MessageAgentThought)
-            .filter_by(message_id=message.id)
-            .order_by(MessageAgentThought.position.asc())
-            .all()
-        )
-
-        if agent_thoughts:
-            # Agent-Chat mode: merge MessageAgentThought records
-            content_pos = 0
-            cleaned_answer_parts: list[str] = []
-            for thought in agent_thoughts:
-                # Add thought/reasoning
-                if thought.thought:
-                    reasoning_text = thought.thought
-                    if "<think" in reasoning_text.lower():
-                        clean_text, extracted_reasoning = self._split_reasoning_from_answer(reasoning_text)
-                        if extracted_reasoning:
-                            reasoning_text = extracted_reasoning
-                            thought.thought = clean_text or extracted_reasoning
-                    reasoning_list.append(reasoning_text)
-                    sequence.append({"type": "reasoning", "index": len(reasoning_list) - 1})
-
-                # Add tool calls
-                if thought.tool:
-                    tool_calls_list.append(
-                        {
-                            "name": thought.tool,
-                            "arguments": thought.tool_input or "",
-                            "result": thought.observation or "",
-                        }
-                    )
-                    sequence.append({"type": "tool_call", "index": len(tool_calls_list) - 1})
-
-                # Add answer content if present
-                if thought.answer:
-                    content_text = thought.answer
-                    if "<think" in content_text.lower():
-                        clean_answer, extracted_reasoning = self._split_reasoning_from_answer(content_text)
-                        if extracted_reasoning:
-                            reasoning_list.append(extracted_reasoning)
-                            sequence.append({"type": "reasoning", "index": len(reasoning_list) - 1})
-                        content_text = clean_answer
-                        thought.answer = clean_answer or content_text
-
-                    if content_text:
-                        start = content_pos
-                        end = content_pos + len(content_text)
-                        sequence.append({"type": "content", "start": start, "end": end})
-                        content_pos = end
-                        cleaned_answer_parts.append(content_text)
-
-            if cleaned_answer_parts:
-                merged_answer = "".join(cleaned_answer_parts)
-                message.answer = merged_answer
-                llm_result.message.content = merged_answer
-        else:
-            # Completion/Chat mode: use reasoning_content from llm_result
-            reasoning_content = llm_result.reasoning_content
-            if not reasoning_content and answer:
-                # Extract reasoning from <think> blocks and clean the final answer
-                clean_answer, reasoning_content = self._split_reasoning_from_answer(answer)
-                if reasoning_content:
-                    answer = clean_answer
-                    llm_result.message.content = clean_answer
-                    llm_result.reasoning_content = reasoning_content
-                    message.answer = clean_answer
-            if reasoning_content:
-                reasoning_list = [reasoning_content]
-                # Content comes first, then reasoning
-                if answer:
-                    sequence.append({"type": "content", "start": 0, "end": len(answer)})
-                sequence.append({"type": "reasoning", "index": 0})
-
-        # Only save if there's meaningful generation detail
-        if not reasoning_list and not tool_calls_list:
-            return
-
-        # Check if generation detail already exists
-        existing = session.query(LLMGenerationDetail).filter_by(message_id=message.id).first()
-
-        if existing:
-            existing.reasoning_content = json.dumps(reasoning_list) if reasoning_list else None
-            existing.tool_calls = json.dumps(tool_calls_list) if tool_calls_list else None
-            existing.sequence = json.dumps(sequence) if sequence else None
-        else:
-            generation_detail = LLMGenerationDetail(
-                tenant_id=self._application_generate_entity.app_config.tenant_id,
-                app_id=self._application_generate_entity.app_config.app_id,
-                message_id=message.id,
-                reasoning_content=json.dumps(reasoning_list) if reasoning_list else None,
-                tool_calls=json.dumps(tool_calls_list) if tool_calls_list else None,
-                sequence=json.dumps(sequence) if sequence else None,
-            )
-            session.add(generation_detail)
-
-    @classmethod
-    def _split_reasoning_from_answer(cls, text: str) -> tuple[str, str]:
-        """
-        Extract reasoning segments from <think> blocks and return (clean_text, reasoning).
-        """
-        matches = cls._THINK_PATTERN.findall(text)
-        reasoning_content = "\n".join(match.strip() for match in matches) if matches else ""
-
-        clean_text = cls._THINK_PATTERN.sub("", text)
-        clean_text = re.sub(r"\n\s*\n", "\n\n", clean_text).strip()
-
-        return clean_text, reasoning_content or ""
-
    def _handle_stop(self, event: QueueStopEvent):
        """
        Handle stop.
@@ -232,31 +232,15 @@ class MessageCycleManager:
        answer: str,
        message_id: str,
        from_variable_selector: list[str] | None = None,
-        chunk_type: str | None = None,
-        tool_call_id: str | None = None,
-        tool_name: str | None = None,
-        tool_arguments: str | None = None,
-        tool_files: list[str] | None = None,
-        tool_error: str | None = None,
-        tool_elapsed_time: float | None = None,
-        tool_icon: str | dict | None = None,
-        tool_icon_dark: str | dict | None = None,
        event_type: StreamEvent | None = None,
    ) -> MessageStreamResponse:
        """
        Message to stream response.
        :param answer: answer
        :param message_id: message id
-        :param from_variable_selector: from variable selector
-        :param chunk_type: type of the chunk (text, function_call, tool_result, thought)
-        :param tool_call_id: unique identifier for this tool call
-        :param tool_name: name of the tool being called
-        :param tool_arguments: accumulated tool arguments JSON
-        :param tool_files: file IDs produced by tool
-        :param tool_error: error message if tool failed
        :return:
        """
-        response = MessageStreamResponse(
+        return MessageStreamResponse(
            task_id=self._application_generate_entity.task_id,
            id=message_id,
            answer=answer,
@@ -264,35 +248,6 @@ class MessageCycleManager:
            event=event_type or StreamEvent.MESSAGE,
        )

-        if chunk_type:
-            response = response.model_copy(update={"chunk_type": chunk_type})
-
-        if chunk_type == "tool_call":
-            response = response.model_copy(
-                update={
-                    "tool_call_id": tool_call_id,
-                    "tool_name": tool_name,
-                    "tool_arguments": tool_arguments,
-                    "tool_icon": tool_icon,
-                    "tool_icon_dark": tool_icon_dark,
-                }
-            )
-        elif chunk_type == "tool_result":
-            response = response.model_copy(
-                update={
-                    "tool_call_id": tool_call_id,
-                    "tool_name": tool_name,
-                    "tool_arguments": tool_arguments,
-                    "tool_files": tool_files,
-                    "tool_error": tool_error,
-                    "tool_elapsed_time": tool_elapsed_time,
-                    "tool_icon": tool_icon,
-                    "tool_icon_dark": tool_icon_dark,
-                }
-            )
-
-        return response
-
    def message_replace_to_stream_response(self, answer: str, reason: str = "") -> MessageReplaceStreamResponse:
        """
        Message replace to stream response.
@@ -5,6 +5,7 @@ from sqlalchemy import select

 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.entities.app_invoke_entities import InvokeFrom
+from core.app.entities.queue_entities import QueueRetrieverResourcesEvent
 from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.rag.index_processor.constant.index_type import IndexStructureType
 from core.rag.models.document import Document
@@ -89,8 +90,6 @@ class DatasetIndexToolCallbackHandler:
    # TODO(-LAN-): Improve type check
    def return_retriever_resource_info(self, resource: Sequence[RetrievalSourceMetadata]):
        """Handle return_retriever_resource_info."""
-        from core.app.entities.queue_entities import QueueRetrieverResourcesEvent
-
        self._queue_manager.publish(
            QueueRetrieverResourcesEvent(retriever_resources=resource), PublishFrom.APPLICATION_MANAGER
        )
@@ -33,6 +33,10 @@ class MaxRetriesExceededError(ValueError):
    pass


+request_error = httpx.RequestError
+max_retries_exceeded_error = MaxRetriesExceededError
+
+
 def _create_proxy_mounts() -> dict[str, httpx.HTTPTransport]:
    return {
        "http://": httpx.HTTPTransport(
@@ -56,6 +56,10 @@ class HostingConfiguration:
        self.provider_map[f"{DEFAULT_PLUGIN_ID}/minimax/minimax"] = self.init_minimax()
        self.provider_map[f"{DEFAULT_PLUGIN_ID}/spark/spark"] = self.init_spark()
        self.provider_map[f"{DEFAULT_PLUGIN_ID}/zhipuai/zhipuai"] = self.init_zhipuai()
+        self.provider_map[f"{DEFAULT_PLUGIN_ID}/gemini/google"] = self.init_gemini()
+        self.provider_map[f"{DEFAULT_PLUGIN_ID}/x/x"] = self.init_xai()
+        self.provider_map[f"{DEFAULT_PLUGIN_ID}/deepseek/deepseek"] = self.init_deepseek()
+        self.provider_map[f"{DEFAULT_PLUGIN_ID}/tongyi/tongyi"] = self.init_tongyi()

        self.moderation_config = self.init_moderation_config()

@@ -128,7 +132,7 @@ class HostingConfiguration:
        quotas: list[HostingQuota] = []

        if dify_config.HOSTED_OPENAI_TRIAL_ENABLED:
-            hosted_quota_limit = dify_config.HOSTED_OPENAI_QUOTA_LIMIT
+            hosted_quota_limit = 0
            trial_models = self.parse_restrict_models_from_env("HOSTED_OPENAI_TRIAL_MODELS")
            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trial_models)
            quotas.append(trial_quota)
@@ -156,18 +160,49 @@ class HostingConfiguration:
            quota_unit=quota_unit,
        )

-    @staticmethod
-    def init_anthropic() -> HostingProvider:
-        quota_unit = QuotaUnit.TOKENS
+    def init_gemini(self) -> HostingProvider:
+        quota_unit = QuotaUnit.CREDITS
+        quotas: list[HostingQuota] = []
+
+        if dify_config.HOSTED_GEMINI_TRIAL_ENABLED:
+            hosted_quota_limit = 0
+            trial_models = self.parse_restrict_models_from_env("HOSTED_GEMINI_TRIAL_MODELS")
+            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trial_models)
+            quotas.append(trial_quota)
+
+        if dify_config.HOSTED_GEMINI_PAID_ENABLED:
+            paid_models = self.parse_restrict_models_from_env("HOSTED_GEMINI_PAID_MODELS")
+            paid_quota = PaidHostingQuota(restrict_models=paid_models)
+            quotas.append(paid_quota)
+
+        if len(quotas) > 0:
+            credentials = {
+                "google_api_key": dify_config.HOSTED_GEMINI_API_KEY,
+            }
+
+            if dify_config.HOSTED_GEMINI_API_BASE:
+                credentials["google_base_url"] = dify_config.HOSTED_GEMINI_API_BASE
+
+            return HostingProvider(enabled=True, credentials=credentials, quota_unit=quota_unit, quotas=quotas)
+
+        return HostingProvider(
+            enabled=False,
+            quota_unit=quota_unit,
+        )
+
+    def init_anthropic(self) -> HostingProvider:
+        quota_unit = QuotaUnit.CREDITS
        quotas: list[HostingQuota] = []

        if dify_config.HOSTED_ANTHROPIC_TRIAL_ENABLED:
-            hosted_quota_limit = dify_config.HOSTED_ANTHROPIC_QUOTA_LIMIT
-            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit)
+            hosted_quota_limit = 0
+            trail_models = self.parse_restrict_models_from_env("HOSTED_ANTHROPIC_TRIAL_MODELS")
+            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trail_models)
            quotas.append(trial_quota)

        if dify_config.HOSTED_ANTHROPIC_PAID_ENABLED:
-            paid_quota = PaidHostingQuota()
+            paid_models = self.parse_restrict_models_from_env("HOSTED_ANTHROPIC_PAID_MODELS")
+            paid_quota = PaidHostingQuota(restrict_models=paid_models)
            quotas.append(paid_quota)

        if len(quotas) > 0:
@@ -185,6 +220,94 @@ class HostingConfiguration:
            quota_unit=quota_unit,
        )

+    def init_tongyi(self) -> HostingProvider:
+        quota_unit = QuotaUnit.CREDITS
+        quotas: list[HostingQuota] = []
+
+        if dify_config.HOSTED_TONGYI_TRIAL_ENABLED:
+            hosted_quota_limit = 0
+            trail_models = self.parse_restrict_models_from_env("HOSTED_TONGYI_TRIAL_MODELS")
+            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trail_models)
+            quotas.append(trial_quota)
+
+        if dify_config.HOSTED_TONGYI_PAID_ENABLED:
+            paid_models = self.parse_restrict_models_from_env("HOSTED_TONGYI_PAID_MODELS")
+            paid_quota = PaidHostingQuota(restrict_models=paid_models)
+            quotas.append(paid_quota)
+
+        if len(quotas) > 0:
+            credentials = {
+                "dashscope_api_key": dify_config.HOSTED_TONGYI_API_KEY,
+                "use_international_endpoint": dify_config.HOSTED_TONGYI_USE_INTERNATIONAL_ENDPOINT,
+            }
+
+            return HostingProvider(enabled=True, credentials=credentials, quota_unit=quota_unit, quotas=quotas)
+
+        return HostingProvider(
+            enabled=False,
+            quota_unit=quota_unit,
+        )
+
+    def init_xai(self) -> HostingProvider:
+        quota_unit = QuotaUnit.CREDITS
+        quotas: list[HostingQuota] = []
+
+        if dify_config.HOSTED_XAI_TRIAL_ENABLED:
+            hosted_quota_limit = 0
+            trail_models = self.parse_restrict_models_from_env("HOSTED_XAI_TRIAL_MODELS")
+            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trail_models)
+            quotas.append(trial_quota)
+
+        if dify_config.HOSTED_XAI_PAID_ENABLED:
+            paid_models = self.parse_restrict_models_from_env("HOSTED_XAI_PAID_MODELS")
+            paid_quota = PaidHostingQuota(restrict_models=paid_models)
+            quotas.append(paid_quota)
+
+        if len(quotas) > 0:
+            credentials = {
+                "api_key": dify_config.HOSTED_XAI_API_KEY,
+            }
+
+            if dify_config.HOSTED_XAI_API_BASE:
+                credentials["endpoint_url"] = dify_config.HOSTED_XAI_API_BASE
+
+            return HostingProvider(enabled=True, credentials=credentials, quota_unit=quota_unit, quotas=quotas)
+
+        return HostingProvider(
+            enabled=False,
+            quota_unit=quota_unit,
+        )
+
+    def init_deepseek(self) -> HostingProvider:
+        quota_unit = QuotaUnit.CREDITS
+        quotas: list[HostingQuota] = []
+
+        if dify_config.HOSTED_DEEPSEEK_TRIAL_ENABLED:
+            hosted_quota_limit = 0
+            trail_models = self.parse_restrict_models_from_env("HOSTED_DEEPSEEK_TRIAL_MODELS")
+            trial_quota = TrialHostingQuota(quota_limit=hosted_quota_limit, restrict_models=trail_models)
+            quotas.append(trial_quota)
+
+        if dify_config.HOSTED_DEEPSEEK_PAID_ENABLED:
+            paid_models = self.parse_restrict_models_from_env("HOSTED_DEEPSEEK_PAID_MODELS")
+            paid_quota = PaidHostingQuota(restrict_models=paid_models)
+            quotas.append(paid_quota)
+
+        if len(quotas) > 0:
+            credentials = {
+                "api_key": dify_config.HOSTED_DEEPSEEK_API_KEY,
+            }
+
+            if dify_config.HOSTED_DEEPSEEK_API_BASE:
+                credentials["endpoint_url"] = dify_config.HOSTED_DEEPSEEK_API_BASE
+
+            return HostingProvider(enabled=True, credentials=credentials, quota_unit=quota_unit, quotas=quotas)
+
+        return HostingProvider(
+            enabled=False,
+            quota_unit=quota_unit,
+        )
+
    @staticmethod
    def init_minimax() -> HostingProvider:
        quota_unit = QuotaUnit.TOKENS
@@ -1,8 +1,8 @@
 import json
 import logging
 import re
-from collections.abc import Sequence
-from typing import Protocol, cast
+from collections.abc import Mapping, Sequence
+from typing import Any, Protocol, cast

 import json_repair

@@ -398,6 +398,488 @@ class LLMGenerator:
            logger.exception("Failed to invoke LLM model, model: %s", model_config.get("name"))
            return {"output": "", "error": f"An unexpected error occurred: {str(e)}"}

+    @classmethod
+    def generate_with_context(
+        cls,
+        tenant_id: str,
+        workflow_id: str,
+        node_id: str,
+        parameter_name: str,
+        language: str,
+        prompt_messages: list[PromptMessage],
+        model_config: dict,
+    ) -> dict:
+        """
+        Generate extractor code node based on conversation context.
+
+        Args:
+            tenant_id: Tenant/workspace ID
+            workflow_id: Workflow ID
+            node_id: Current tool/llm node ID
+            parameter_name: Parameter name to generate code for
+            language: Code language (python3/javascript)
+            prompt_messages: Multi-turn conversation history (last message is instruction)
+            model_config: Model configuration (provider, name, completion_params)
+
+        Returns:
+            dict with CodeNodeData format:
+            - variables: Input variable selectors
+            - code_language: Code language
+            - code: Generated code
+            - outputs: Output definitions
+            - message: Explanation
+            - error: Error message if any
+        """
+        from sqlalchemy import select
+        from sqlalchemy.orm import Session
+
+        from services.workflow_service import WorkflowService
+
+        # Get workflow
+        with Session(db.engine) as session:
+            stmt = select(App).where(App.id == workflow_id)
+            app = session.scalar(stmt)
+            if not app:
+                return cls._error_response(f"App {workflow_id} not found")
+
+            workflow = WorkflowService().get_draft_workflow(app_model=app)
+            if not workflow:
+                return cls._error_response(f"Workflow for app {workflow_id} not found")
+
+        # Get upstream nodes via edge backtracking
+        upstream_nodes = cls._get_upstream_nodes(workflow.graph_dict, node_id)
+
+        # Get current node info
+        current_node = cls._get_node_by_id(workflow.graph_dict, node_id)
+        if not current_node:
+            return cls._error_response(f"Node {node_id} not found")
+
+        # Get parameter info
+        parameter_info = cls._get_parameter_info(
+            tenant_id=tenant_id,
+            node_data=current_node.get("data", {}),
+            parameter_name=parameter_name,
+        )
+
+        # Build system prompt
+        system_prompt = cls._build_extractor_system_prompt(
+            upstream_nodes=upstream_nodes,
+            current_node=current_node,
+            parameter_info=parameter_info,
+            language=language,
+        )
+
+        # Construct complete prompt_messages with system prompt
+        complete_messages: list[PromptMessage] = [
+            SystemPromptMessage(content=system_prompt),
+            *prompt_messages,
+        ]
+
+        from core.llm_generator.output_parser.structured_output import invoke_llm_with_structured_output
+
+        # Get model instance and schema
+        provider = model_config.get("provider", "")
+        model_name = model_config.get("name", "")
+        model_instance = ModelManager().get_model_instance(
+            tenant_id=tenant_id,
+            model_type=ModelType.LLM,
+            provider=provider,
+            model=model_name,
+        )
+
+        model_schema = model_instance.model_type_instance.get_model_schema(model_name, model_instance.credentials)
+        if not model_schema:
+            return cls._error_response(f"Model schema not found for {model_name}")
+
+        model_parameters = model_config.get("completion_params", {})
+        json_schema = cls._get_code_node_json_schema()
+
+        try:
+            response = invoke_llm_with_structured_output(
+                provider=provider,
+                model_schema=model_schema,
+                model_instance=model_instance,
+                prompt_messages=complete_messages,
+                json_schema=json_schema,
+                model_parameters=model_parameters,
+                stream=False,
+                tenant_id=tenant_id,
+            )
+
+            return cls._parse_code_node_output(
+                response.structured_output, language, parameter_info.get("type", "string")
+            )
+
+        except InvokeError as e:
+            return cls._error_response(str(e))
+        except Exception as e:
+            logger.exception("Failed to generate with context, model: %s", model_config.get("name"))
+            return cls._error_response(f"An unexpected error occurred: {str(e)}")
+
+    @classmethod
+    def _error_response(cls, error: str) -> dict:
+        """Return error response in CodeNodeData format."""
+        return {
+            "variables": [],
+            "code_language": "python3",
+            "code": "",
+            "outputs": {},
+            "message": "",
+            "error": error,
+        }
+
+    @classmethod
+    def generate_suggested_questions(
+        cls,
+        tenant_id: str,
+        workflow_id: str,
+        node_id: str,
+        parameter_name: str,
+        language: str,
+        model_config: dict | None = None,
+    ) -> dict:
+        """
+        Generate suggested questions for context generation.
+
+        Returns dict with questions array and error field.
+        """
+        from sqlalchemy import select
+        from sqlalchemy.orm import Session
+
+        from core.llm_generator.output_parser.structured_output import invoke_llm_with_structured_output
+        from services.workflow_service import WorkflowService
+
+        # Get workflow context (reuse existing logic)
+        with Session(db.engine) as session:
+            stmt = select(App).where(App.id == workflow_id)
+            app = session.scalar(stmt)
+            if not app:
+                return {"questions": [], "error": f"App {workflow_id} not found"}
+
+            workflow = WorkflowService().get_draft_workflow(app_model=app)
+            if not workflow:
+                return {"questions": [], "error": f"Workflow for app {workflow_id} not found"}
+
+        upstream_nodes = cls._get_upstream_nodes(workflow.graph_dict, node_id)
+        current_node = cls._get_node_by_id(workflow.graph_dict, node_id)
+        if not current_node:
+            return {"questions": [], "error": f"Node {node_id} not found"}
+
+        parameter_info = cls._get_parameter_info(
+            tenant_id=tenant_id,
+            node_data=current_node.get("data", {}),
+            parameter_name=parameter_name,
+        )
+
+        # Build prompt
+        system_prompt = cls._build_suggested_questions_prompt(
+            upstream_nodes=upstream_nodes,
+            current_node=current_node,
+            parameter_info=parameter_info,
+            language=language,
+        )
+
+        prompt_messages: list[PromptMessage] = [
+            SystemPromptMessage(content=system_prompt),
+        ]
+
+        # Get model instance - use default if model_config not provided
+        model_manager = ModelManager()
+        if model_config:
+            provider = model_config.get("provider", "")
+            model_name = model_config.get("name", "")
+            model_instance = model_manager.get_model_instance(
+                tenant_id=tenant_id,
+                model_type=ModelType.LLM,
+                provider=provider,
+                model=model_name,
+            )
+        else:
+            model_instance = model_manager.get_default_model_instance(
+                tenant_id=tenant_id,
+                model_type=ModelType.LLM,
+            )
+            model_name = model_instance.model
+
+        model_schema = model_instance.model_type_instance.get_model_schema(model_name, model_instance.credentials)
+        if not model_schema:
+            return {"questions": [], "error": f"Model schema not found for {model_name}"}
+
+        completion_params = model_config.get("completion_params", {}) if model_config else {}
+        model_parameters = {**completion_params, "max_tokens": 256}
+        json_schema = cls._get_suggested_questions_json_schema()
+
+        try:
+            response = invoke_llm_with_structured_output(
+                provider=model_instance.provider,
+                model_schema=model_schema,
+                model_instance=model_instance,
+                prompt_messages=prompt_messages,
+                json_schema=json_schema,
+                model_parameters=model_parameters,
+                stream=False,
+                tenant_id=tenant_id,
+            )
+
+            questions = response.structured_output.get("questions", []) if response.structured_output else []
+            return {"questions": questions, "error": ""}
+
+        except InvokeError as e:
+            return {"questions": [], "error": str(e)}
+        except Exception as e:
+            logger.exception("Failed to generate suggested questions, model: %s", model_name)
+            return {"questions": [], "error": f"An unexpected error occurred: {str(e)}"}
+
+    @classmethod
+    def _build_suggested_questions_prompt(
+        cls,
+        upstream_nodes: list[dict],
+        current_node: dict,
+        parameter_info: dict,
+        language: str = "English",
+    ) -> str:
+        """Build minimal prompt for suggested questions generation."""
+        # Simplify upstream nodes to reduce tokens
+        sources = [f"{n['title']}({','.join(n.get('outputs', {}).keys())})" for n in upstream_nodes[:5]]
+        param_type = parameter_info.get("type", "string")
+        param_desc = parameter_info.get("description", "")[:100]
+
+        return f"""Suggest 3 code generation questions for extracting data.
+Sources: {", ".join(sources)}
+Target: {parameter_info.get("name")}({param_type}) - {param_desc}
+Output 3 short, practical questions in {language}."""
+
+    @classmethod
+    def _get_suggested_questions_json_schema(cls) -> dict:
+        """Return JSON Schema for suggested questions."""
+        return {
+            "type": "object",
+            "properties": {
+                "questions": {
+                    "type": "array",
+                    "items": {"type": "string"},
+                    "minItems": 3,
+                    "maxItems": 3,
+                    "description": "3 suggested questions",
+                },
+            },
+            "required": ["questions"],
+        }
+
+    @classmethod
+    def _get_code_node_json_schema(cls) -> dict:
+        """Return JSON Schema for structured output."""
+        return {
+            "type": "object",
+            "properties": {
+                "variables": {
+                    "type": "array",
+                    "items": {
+                        "type": "object",
+                        "properties": {
+                            "variable": {"type": "string", "description": "Variable name in code"},
+                            "value_selector": {
+                                "type": "array",
+                                "items": {"type": "string"},
+                                "description": "Path like [node_id, output_name]",
+                            },
+                        },
+                        "required": ["variable", "value_selector"],
+                    },
+                },
+                "code": {"type": "string", "description": "Generated code with main function"},
+                "outputs": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "object",
+                        "properties": {"type": {"type": "string"}},
+                    },
+                    "description": "Output definitions, key is output name",
+                },
+                "explanation": {"type": "string", "description": "Brief explanation of the code"},
+            },
+            "required": ["variables", "code", "outputs", "explanation"],
+        }
+
+    @classmethod
+    def _get_upstream_nodes(cls, graph_dict: Mapping[str, Any], node_id: str) -> list[dict]:
+        """
+        Get all upstream nodes via edge backtracking.
+
+        Traverses the graph backwards from node_id to collect all reachable nodes.
+        """
+        from collections import defaultdict
+
+        nodes = {n["id"]: n for n in graph_dict.get("nodes", [])}
+        edges = graph_dict.get("edges", [])
+
+        # Build reverse adjacency list
+        reverse_adj: dict[str, list[str]] = defaultdict(list)
+        for edge in edges:
+            reverse_adj[edge["target"]].append(edge["source"])
+
+        # BFS to find all upstream nodes
+        visited: set[str] = set()
+        queue = [node_id]
+        upstream: list[dict] = []
+
+        while queue:
+            current = queue.pop(0)
+            for source in reverse_adj.get(current, []):
+                if source not in visited:
+                    visited.add(source)
+                    queue.append(source)
+                    if source in nodes:
+                        upstream.append(cls._extract_node_info(nodes[source]))
+
+        return upstream
+
+    @classmethod
+    def _get_node_by_id(cls, graph_dict: Mapping[str, Any], node_id: str) -> dict | None:
+        """Get node by ID from graph."""
+        for node in graph_dict.get("nodes", []):
+            if node["id"] == node_id:
+                return node
+        return None
+
+    @classmethod
+    def _extract_node_info(cls, node: dict) -> dict:
+        """Extract minimal node info with outputs based on node type."""
+        node_type = node["data"]["type"]
+        node_data = node.get("data", {})
+
+        # Build outputs based on node type (only type, no description to reduce tokens)
+        outputs: dict[str, str] = {}
+        match node_type:
+            case "start":
+                for var in node_data.get("variables", []):
+                    name = var.get("variable", var.get("name", ""))
+                    outputs[name] = var.get("type", "string")
+            case "llm":
+                outputs["text"] = "string"
+            case "code":
+                for name, output in node_data.get("outputs", {}).items():
+                    outputs[name] = output.get("type", "string")
+            case "http-request":
+                outputs = {"body": "string", "status_code": "number", "headers": "object"}
+            case "knowledge-retrieval":
+                outputs["result"] = "array[object]"
+            case "tool":
+                outputs = {"text": "string", "json": "object"}
+            case _:
+                outputs["output"] = "string"
+
+        info: dict = {
+            "id": node["id"],
+            "title": node_data.get("title", node["id"]),
+            "outputs": outputs,
+        }
+        # Only include description if not empty
+        desc = node_data.get("desc", "")
+        if desc:
+            info["desc"] = desc
+
+        return info
+
+    @classmethod
+    def _get_parameter_info(cls, tenant_id: str, node_data: dict, parameter_name: str) -> dict:
+        """Get parameter info from tool schema using ToolManager."""
+        default_info = {"name": parameter_name, "type": "string", "description": ""}
+
+        if node_data.get("type") != "tool":
+            return default_info
+
+        try:
+            from core.app.entities.app_invoke_entities import InvokeFrom
+            from core.tools.entities.tool_entities import ToolProviderType
+            from core.tools.tool_manager import ToolManager
+
+            provider_type_str = node_data.get("provider_type", "")
+            provider_type = ToolProviderType(provider_type_str) if provider_type_str else ToolProviderType.BUILT_IN
+
+            tool_runtime = ToolManager.get_tool_runtime(
+                provider_type=provider_type,
+                provider_id=node_data.get("provider_id", ""),
+                tool_name=node_data.get("tool_name", ""),
+                tenant_id=tenant_id,
+                invoke_from=InvokeFrom.DEBUGGER,
+            )
+
+            parameters = tool_runtime.get_merged_runtime_parameters()
+            for param in parameters:
+                if param.name == parameter_name:
+                    return {
+                        "name": param.name,
+                        "type": param.type.value if hasattr(param.type, "value") else str(param.type),
+                        "description": param.llm_description
+                        or (param.human_description.en_US if param.human_description else ""),
+                        "required": param.required,
+                    }
+        except Exception as e:
+            logger.debug("Failed to get parameter info from ToolManager: %s", e)
+
+        return default_info
+
+    @classmethod
+    def _build_extractor_system_prompt(
+        cls,
+        upstream_nodes: list[dict],
+        current_node: dict,
+        parameter_info: dict,
+        language: str,
+    ) -> str:
+        """Build system prompt for extractor code generation."""
+        upstream_json = json.dumps(upstream_nodes, indent=2, ensure_ascii=False)
+        param_type = parameter_info.get("type", "string")
+        return f"""You are a code generator for workflow automation.
+
+Generate {language} code to extract/transform upstream node outputs for the target parameter.
+
+## Upstream Nodes
+{upstream_json}
+
+## Target
+Node: {current_node["data"].get("title", current_node["id"])}
+Parameter: {parameter_info.get("name")} ({param_type}) - {parameter_info.get("description", "")}
+
+## Requirements
+- Write a main function that returns type: {param_type}
+- Use value_selector format: ["node_id", "output_name"]
+"""
+
+    @classmethod
+    def _parse_code_node_output(cls, content: Mapping[str, Any] | None, language: str, parameter_type: str) -> dict:
+        """
+        Parse structured output to CodeNodeData format.
+
+        Args:
+            content: Structured output dict from invoke_llm_with_structured_output
+            language: Code language
+            parameter_type: Expected parameter type
+
+        Returns dict with variables, code_language, code, outputs, message, error.
+        """
+        if content is None:
+            return cls._error_response("Empty or invalid response from LLM")
+
+        # Validate and normalize variables
+        variables = [
+            {"variable": v.get("variable", ""), "value_selector": v.get("value_selector", [])}
+            for v in content.get("variables", [])
+            if isinstance(v, dict)
+        ]
+
+        outputs = content.get("outputs", {"result": {"type": parameter_type}})
+
+        return {
+            "variables": variables,
+            "code_language": language,
+            "code": content.get("code", ""),
+            "outputs": outputs,
+            "message": content.get("explanation", ""),
+            "error": "",
+        }
+
    @staticmethod
    def instruction_modify_legacy(
        tenant_id: str, flow_id: str, current: str, instruction: str, model_config: dict, ideal_output: str | None
@@ -0,0 +1,188 @@
+"""
+File reference detection and conversion for structured output.
+
+This module provides utilities to:
+1. Detect file reference fields in JSON Schema (format: "dify-file-ref")
+2. Convert file ID strings to File objects after LLM returns
+"""
+
+import uuid
+from collections.abc import Mapping
+from typing import Any
+
+from core.file import File
+from core.variables.segments import ArrayFileSegment, FileSegment
+from factories.file_factory import build_from_mapping
+
+FILE_REF_FORMAT = "dify-file-ref"
+
+
+def is_file_ref_property(schema: dict) -> bool:
+    """Check if a schema property is a file reference."""
+    return schema.get("type") == "string" and schema.get("format") == FILE_REF_FORMAT
+
+
+def detect_file_ref_fields(schema: Mapping[str, Any], path: str = "") -> list[str]:
+    """
+    Recursively detect file reference fields in schema.
+
+    Args:
+        schema: JSON Schema to analyze
+        path: Current path in the schema (used for recursion)
+
+    Returns:
+        List of JSON paths containing file refs, e.g., ["image_id", "files[*]"]
+    """
+    file_ref_paths: list[str] = []
+    schema_type = schema.get("type")
+
+    if schema_type == "object":
+        for prop_name, prop_schema in schema.get("properties", {}).items():
+            current_path = f"{path}.{prop_name}" if path else prop_name
+
+            if is_file_ref_property(prop_schema):
+                file_ref_paths.append(current_path)
+            elif isinstance(prop_schema, dict):
+                file_ref_paths.extend(detect_file_ref_fields(prop_schema, current_path))
+
+    elif schema_type == "array":
+        items_schema = schema.get("items", {})
+        array_path = f"{path}[*]" if path else "[*]"
+
+        if is_file_ref_property(items_schema):
+            file_ref_paths.append(array_path)
+        elif isinstance(items_schema, dict):
+            file_ref_paths.extend(detect_file_ref_fields(items_schema, array_path))
+
+    return file_ref_paths
+
+
+def convert_file_refs_in_output(
+    output: Mapping[str, Any],
+    json_schema: Mapping[str, Any],
+    tenant_id: str,
+) -> dict[str, Any]:
+    """
+    Convert file ID strings to File objects based on schema.
+
+    Args:
+        output: The structured_output from LLM result
+        json_schema: The original JSON schema (to detect file ref fields)
+        tenant_id: Tenant ID for file lookup
+
+    Returns:
+        Output with file references converted to File objects
+    """
+    file_ref_paths = detect_file_ref_fields(json_schema)
+    if not file_ref_paths:
+        return dict(output)
+
+    result = _deep_copy_dict(output)
+
+    for path in file_ref_paths:
+        _convert_path_in_place(result, path.split("."), tenant_id)
+
+    return result
+
+
+def _deep_copy_dict(obj: Mapping[str, Any]) -> dict[str, Any]:
+    """Deep copy a mapping to a mutable dict."""
+    result: dict[str, Any] = {}
+    for key, value in obj.items():
+        if isinstance(value, Mapping):
+            result[key] = _deep_copy_dict(value)
+        elif isinstance(value, list):
+            result[key] = [_deep_copy_dict(item) if isinstance(item, Mapping) else item for item in value]
+        else:
+            result[key] = value
+    return result
+
+
+def _convert_path_in_place(obj: dict, path_parts: list[str], tenant_id: str) -> None:
+    """Convert file refs at the given path in place, wrapping in Segment types."""
+    if not path_parts:
+        return
+
+    current = path_parts[0]
+    remaining = path_parts[1:]
+
+    # Handle array notation like "files[*]"
+    if current.endswith("[*]"):
+        key = current[:-3] if current != "[*]" else None
+        target = obj.get(key) if key else obj
+
+        if isinstance(target, list):
+            if remaining:
+                # Nested array with remaining path - recurse into each item
+                for item in target:
+                    if isinstance(item, dict):
+                        _convert_path_in_place(item, remaining, tenant_id)
+            else:
+                # Array of file IDs - convert all and wrap in ArrayFileSegment
+                files: list[File] = []
+                for item in target:
+                    file = _convert_file_id(item, tenant_id)
+                    if file is not None:
+                        files.append(file)
+                # Replace the array with ArrayFileSegment
+                if key:
+                    obj[key] = ArrayFileSegment(value=files)
+        return
+
+    if not remaining:
+        # Leaf node - convert the value and wrap in FileSegment
+        if current in obj:
+            file = _convert_file_id(obj[current], tenant_id)
+            if file is not None:
+                obj[current] = FileSegment(value=file)
+            else:
+                obj[current] = None
+    else:
+        # Recurse into nested object
+        if current in obj and isinstance(obj[current], dict):
+            _convert_path_in_place(obj[current], remaining, tenant_id)
+
+
+def _convert_file_id(file_id: Any, tenant_id: str) -> File | None:
+    """
+    Convert a file ID string to a File object.
+
+    Tries multiple file sources in order:
+    1. ToolFile (files generated by tools/workflows)
+    2. UploadFile (files uploaded by users)
+    """
+    if not isinstance(file_id, str):
+        return None
+
+    # Validate UUID format
+    try:
+        uuid.UUID(file_id)
+    except ValueError:
+        return None
+
+    # Try ToolFile first (files generated by tools/workflows)
+    try:
+        return build_from_mapping(
+            mapping={
+                "transfer_method": "tool_file",
+                "tool_file_id": file_id,
+            },
+            tenant_id=tenant_id,
+        )
+    except ValueError:
+        pass
+
+    # Try UploadFile (files uploaded by users)
+    try:
+        return build_from_mapping(
+            mapping={
+                "transfer_method": "local_file",
+                "upload_file_id": file_id,
+            },
+            tenant_id=tenant_id,
+        )
+    except ValueError:
+        pass
+
+    # File not found in any source
+    return None
@@ -8,6 +8,7 @@ import json_repair
 from pydantic import TypeAdapter, ValidationError

 from core.llm_generator.output_parser.errors import OutputParserError
+from core.llm_generator.output_parser.file_ref import convert_file_refs_in_output
 from core.llm_generator.prompts import STRUCTURED_OUTPUT_PROMPT
 from core.model_manager import ModelInstance
 from core.model_runtime.callbacks.base_callback import Callback
@@ -57,6 +58,7 @@ def invoke_llm_with_structured_output(
    stream: Literal[True],
    user: str | None = None,
    callbacks: list[Callback] | None = None,
+    tenant_id: str | None = None,
 ) -> Generator[LLMResultChunkWithStructuredOutput, None, None]: ...
@overload
 def invoke_llm_with_structured_output(
@@ -72,6 +74,7 @@ def invoke_llm_with_structured_output(
    stream: Literal[False],
    user: str | None = None,
    callbacks: list[Callback] | None = None,
+    tenant_id: str | None = None,
 ) -> LLMResultWithStructuredOutput: ...
@overload
 def invoke_llm_with_structured_output(
@@ -87,6 +90,7 @@ def invoke_llm_with_structured_output(
    stream: bool = True,
    user: str | None = None,
    callbacks: list[Callback] | None = None,
+    tenant_id: str | None = None,
 ) -> LLMResultWithStructuredOutput | Generator[LLMResultChunkWithStructuredOutput, None, None]: ...
 def invoke_llm_with_structured_output(
    *,
@@ -101,20 +105,28 @@ def invoke_llm_with_structured_output(
    stream: bool = True,
    user: str | None = None,
    callbacks: list[Callback] | None = None,
+    tenant_id: str | None = None,
 ) -> LLMResultWithStructuredOutput | Generator[LLMResultChunkWithStructuredOutput, None, None]:
    """
-    Invoke large language model with structured output
-    1. This method invokes model_instance.invoke_llm with json_schema
-    2. Try to parse the result as structured output
+    Invoke large language model with structured output.

+    This method invokes model_instance.invoke_llm with json_schema and parses
+    the result as structured output.
+
+    :param provider: model provider name
+    :param model_schema: model schema entity
+    :param model_instance: model instance to invoke
    :param prompt_messages: prompt messages
-    :param json_schema: json schema
+    :param json_schema: json schema for structured output
    :param model_parameters: model parameters
    :param tools: tools for tool calling
    :param stop: stop words
    :param stream: is stream response
    :param user: unique user id
    :param callbacks: callbacks
+    :param tenant_id: tenant ID for file reference conversion. When provided and
+                      json_schema contains file reference fields (format: "dify-file-ref"),
+                      file IDs in the output will be automatically converted to File objects.
    :return: full response or stream response chunk generator result
    """

@@ -153,8 +165,18 @@ def invoke_llm_with_structured_output(
                f"Failed to parse structured output, LLM result is not a string: {llm_result.message.content}"
            )

+        structured_output = _parse_structured_output(llm_result.message.content)
+
+        # Convert file references if tenant_id is provided
+        if tenant_id is not None:
+            structured_output = convert_file_refs_in_output(
+                output=structured_output,
+                json_schema=json_schema,
+                tenant_id=tenant_id,
+            )
+
        return LLMResultWithStructuredOutput(
-            structured_output=_parse_structured_output(llm_result.message.content),
+            structured_output=structured_output,
            model=llm_result.model,
            message=llm_result.message,
            usage=llm_result.usage,
@@ -186,8 +208,18 @@ def invoke_llm_with_structured_output(
                    delta=event.delta,
                )

+            structured_output = _parse_structured_output(result_text)
+
+            # Convert file references if tenant_id is provided
+            if tenant_id is not None:
+                structured_output = convert_file_refs_in_output(
+                    output=structured_output,
+                    json_schema=json_schema,
+                    tenant_id=tenant_id,
+                )
+
            yield LLMResultChunkWithStructuredOutput(
-                structured_output=_parse_structured_output(result_text),
+                structured_output=structured_output,
                model=model_schema.model,
                prompt_messages=prompt_messages,
                system_fingerprint=system_fingerprint,
@@ -0,0 +1,45 @@
+"""Utility functions for LLM generator."""
+
+from core.model_runtime.entities.message_entities import (
+    AssistantPromptMessage,
+    PromptMessage,
+    PromptMessageRole,
+    SystemPromptMessage,
+    ToolPromptMessage,
+    UserPromptMessage,
+)
+
+
+def deserialize_prompt_messages(messages: list[dict]) -> list[PromptMessage]:
+    """
+    Deserialize list of dicts to list[PromptMessage].
+
+    Expected format:
+        [
+            {"role": "user", "content": "..."},
+            {"role": "assistant", "content": "..."},
+        ]
+    """
+    result: list[PromptMessage] = []
+    for msg in messages:
+        role = PromptMessageRole.value_of(msg["role"])
+        content = msg.get("content", "")
+
+        match role:
+            case PromptMessageRole.USER:
+                result.append(UserPromptMessage(content=content))
+            case PromptMessageRole.ASSISTANT:
+                result.append(AssistantPromptMessage(content=content))
+            case PromptMessageRole.SYSTEM:
+                result.append(SystemPromptMessage(content=content))
+            case PromptMessageRole.TOOL:
+                result.append(ToolPromptMessage(content=content, tool_call_id=msg.get("tool_call_id", "")))
+
+    return result
+
+
+def serialize_prompt_messages(messages: list[PromptMessage]) -> list[dict]:
+    """
+    Serialize list[PromptMessage] to list of dicts.
+    """
+    return [{"role": msg.role.value, "content": msg.content} for msg in messages]
@@ -0,0 +1,434 @@
+# Memory Module
+
+This module provides memory management for LLM conversations, enabling context retention across dialogue turns.
+
+## Overview
+
+The memory module contains two types of memory implementations:
+
+1. **TokenBufferMemory** - Conversation-level memory (existing)
+2. **NodeTokenBufferMemory** - Node-level memory (to be implemented, **Chatflow only**)
+
+> **Note**: `NodeTokenBufferMemory` is only available in **Chatflow** (advanced-chat mode).
+> This is because it requires both `conversation_id` and `node_id`, which are only present in Chatflow.
+> Standard Workflow mode does not have `conversation_id` and therefore cannot use node-level memory.
+
+```
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                           Memory Architecture                               │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  ┌─────────────────────────────────────────────────────────────────────-┐   │
+│  │                      TokenBufferMemory                               │   │
+│  │  Scope: Conversation                                                 │   │
+│  │  Storage: Database (Message table)                                   │   │
+│  │  Key: conversation_id                                                │   │
+│  └─────────────────────────────────────────────────────────────────────-┘   │
+│                                                                             │
+│  ┌─────────────────────────────────────────────────────────────────────-┐   │
+│  │                    NodeTokenBufferMemory                             │   │
+│  │  Scope: Node within Conversation                                     │   │
+│  │  Storage: Object Storage (JSON file)                                 │   │
+│  │  Key: (app_id, conversation_id, node_id)                             │   │
+│  └─────────────────────────────────────────────────────────────────────-┘   │
+│                                                                             │
+└─────────────────────────────────────────────────────────────────────────────┘
+```
+
+---
+
+## TokenBufferMemory (Existing)
+
+### Purpose
+
+`TokenBufferMemory` retrieves conversation history from the `Message` table and converts it to `PromptMessage` objects for LLM context.
+
+### Key Features
+
+- **Conversation-scoped**: All messages within a conversation are candidates
+- **Thread-aware**: Uses `parent_message_id` to extract only the current thread (supports regeneration scenarios)
+- **Token-limited**: Truncates history to fit within `max_token_limit`
+- **File support**: Handles `MessageFile` attachments (images, documents, etc.)
+
+### Data Flow
+
+```
+Message Table                    TokenBufferMemory              LLM
+     │                                  │                        │
+     │  SELECT * FROM messages          │                        │
+     │  WHERE conversation_id = ?       │                        │
+     │  ORDER BY created_at DESC        │                        │
+     ├─────────────────────────────────▶│                        │
+     │                                  │                        │
+     │                    extract_thread_messages()              │
+     │                                  │                        │
+     │                    build_prompt_message_with_files()      │
+     │                                  │                        │
+     │                    truncate by max_token_limit            │
+     │                                  │                        │
+     │                                  │  Sequence[PromptMessage]
+     │                                  ├───────────────────────▶│
+     │                                  │                        │
+```
+
+### Thread Extraction
+
+When a user regenerates a response, a new thread is created:
+
+```
+Message A (user)
+    └── Message A' (assistant)
+            └── Message B (user)
+                    └── Message B' (assistant)
+            └── Message A'' (assistant, regenerated)  ← New thread
+                    └── Message C (user)
+                            └── Message C' (assistant)
+```
+
+`extract_thread_messages()` traces back from the latest message using `parent_message_id` to get only the current thread: `[A, A'', C, C']`
+
+### Usage
+
+```python
+from core.memory.token_buffer_memory import TokenBufferMemory
+
+memory = TokenBufferMemory(conversation=conversation, model_instance=model_instance)
+history = memory.get_history_prompt_messages(max_token_limit=2000, message_limit=100)
+```
+
+---
+
+## NodeTokenBufferMemory (To Be Implemented)
+
+### Purpose
+
+`NodeTokenBufferMemory` provides **node-scoped memory** within a conversation. Each LLM node in a workflow can maintain its own independent conversation history.
+
+### Use Cases
+
+1. **Multi-LLM Workflows**: Different LLM nodes need separate context
+2. **Iterative Processing**: An LLM node in a loop needs to accumulate context across iterations
+3. **Specialized Agents**: Each agent node maintains its own dialogue history
+
+### Design Decisions
+
+#### Storage: Object Storage for Messages (No New Database Table)
+
+| Aspect                    | Database             | Object Storage     |
+| ------------------------- | -------------------- | ------------------ |
+| Cost                      | High                 | Low                |
+| Query Flexibility         | High                 | Low                |
+| Schema Changes            | Migration required   | None               |
+| Consistency with existing | ConversationVariable | File uploads, logs |
+
+**Decision**: Store message data in object storage, but still use existing database tables for file metadata.
+
+**What is stored in Object Storage:**
+
+- Message content (text)
+- Message metadata (role, token_count, created_at)
+- File references (upload_file_id, tool_file_id, etc.)
+- Thread relationships (message_id, parent_message_id)
+
+**What still requires Database queries:**
+
+- File reconstruction: When reading node memory, file references are used to query
+  `UploadFile` / `ToolFile` tables via `file_factory.build_from_mapping()` to rebuild
+  complete `File` objects with storage_key, mime_type, etc.
+
+**Why this hybrid approach:**
+
+- No database migration required (no new tables)
+- Message data may be large, object storage is cost-effective
+- File metadata is already in database, no need to duplicate
+- Aligns with existing storage patterns (file uploads, logs)
+
+#### Storage Key Format
+
+```
+node_memory/{app_id}/{conversation_id}/{node_id}.json
+```
+
+#### Data Structure
+
+```json
+{
+  "version": 1,
+  "messages": [
+    {
+      "message_id": "msg-001",
+      "parent_message_id": null,
+      "role": "user",
+      "content": "Analyze this image",
+      "files": [
+        {
+          "type": "image",
+          "transfer_method": "local_file",
+          "upload_file_id": "file-uuid-123",
+          "belongs_to": "user"
+        }
+      ],
+      "token_count": 15,
+      "created_at": "2026-01-07T10:00:00Z"
+    },
+    {
+      "message_id": "msg-002",
+      "parent_message_id": "msg-001",
+      "role": "assistant",
+      "content": "This is a landscape image...",
+      "files": [],
+      "token_count": 50,
+      "created_at": "2026-01-07T10:00:01Z"
+    }
+  ]
+}
+```
+
+### Thread Support
+
+Node memory also supports thread extraction (for regeneration scenarios):
+
+```python
+def _extract_thread(
+    self,
+    messages: list[NodeMemoryMessage],
+    current_message_id: str
+) -> list[NodeMemoryMessage]:
+    """
+    Extract messages belonging to the thread of current_message_id.
+    Similar to extract_thread_messages() in TokenBufferMemory.
+    """
+    ...
+```
+
+### File Handling
+
+Files are stored as references (not full metadata):
+
+```python
+class NodeMemoryFile(BaseModel):
+    type: str                        # image, audio, video, document, custom
+    transfer_method: str             # local_file, remote_url, tool_file
+    upload_file_id: str | None       # for local_file
+    tool_file_id: str | None         # for tool_file
+    url: str | None                  # for remote_url
+    belongs_to: str                  # user / assistant
+```
+
+When reading, files are rebuilt using `file_factory.build_from_mapping()`.
+
+### API Design
+
+```python
+class NodeTokenBufferMemory:
+    def __init__(
+        self,
+        app_id: str,
+        conversation_id: str,
+        node_id: str,
+        model_instance: ModelInstance,
+    ):
+        """
+        Initialize node-level memory.
+
+        :param app_id: Application ID
+        :param conversation_id: Conversation ID
+        :param node_id: Node ID in the workflow
+        :param model_instance: Model instance for token counting
+        """
+        ...
+
+    def add_messages(
+        self,
+        message_id: str,
+        parent_message_id: str | None,
+        user_content: str,
+        user_files: Sequence[File],
+        assistant_content: str,
+        assistant_files: Sequence[File],
+    ) -> None:
+        """
+        Append a dialogue turn (user + assistant) to node memory.
+        Call this after LLM node execution completes.
+
+        :param message_id: Current message ID (from Message table)
+        :param parent_message_id: Parent message ID (for thread tracking)
+        :param user_content: User's text input
+        :param user_files: Files attached by user
+        :param assistant_content: Assistant's text response
+        :param assistant_files: Files generated by assistant
+        """
+        ...
+
+    def get_history_prompt_messages(
+        self,
+        current_message_id: str,
+        tenant_id: str,
+        max_token_limit: int = 2000,
+        file_upload_config: FileUploadConfig | None = None,
+    ) -> Sequence[PromptMessage]:
+        """
+        Retrieve history as PromptMessage sequence.
+
+        :param current_message_id: Current message ID (for thread extraction)
+        :param tenant_id: Tenant ID (for file reconstruction)
+        :param max_token_limit: Maximum tokens for history
+        :param file_upload_config: File upload configuration
+        :return: Sequence of PromptMessage for LLM context
+        """
+        ...
+
+    def flush(self) -> None:
+        """
+        Persist buffered changes to object storage.
+        Call this at the end of node execution.
+        """
+        ...
+
+    def clear(self) -> None:
+        """
+        Clear all messages in this node's memory.
+        """
+        ...
+```
+
+### Data Flow
+
+```
+Object Storage                  NodeTokenBufferMemory           LLM Node
+     │                                  │                           │
+     │                                  │◀── get_history_prompt_messages()
+     │  storage.load(key)               │                           │
+     │◀─────────────────────────────────┤                           │
+     │                                  │                           │
+     │  JSON data                       │                           │
+     ├─────────────────────────────────▶│                           │
+     │                                  │                           │
+     │                    _extract_thread()                         │
+     │                                  │                           │
+     │                    _rebuild_files() via file_factory         │
+     │                                  │                           │
+     │                    _build_prompt_messages()                  │
+     │                                  │                           │
+     │                    _truncate_by_tokens()                     │
+     │                                  │                           │
+     │                                  │  Sequence[PromptMessage]  │
+     │                                  ├──────────────────────────▶│
+     │                                  │                           │
+     │                                  │◀── LLM execution complete │
+     │                                  │                           │
+     │                                  │◀── add_messages()         │
+     │                                  │                           │
+     │  storage.save(key, data)         │                           │
+     │◀─────────────────────────────────┤                           │
+     │                                  │                           │
+```
+
+### Integration with LLM Node
+
+```python
+# In LLM Node execution
+
+# 1. Fetch memory based on mode
+if node_data.memory and node_data.memory.mode == MemoryMode.NODE:
+    # Node-level memory (Chatflow only)
+    memory = fetch_node_memory(
+        variable_pool=variable_pool,
+        app_id=app_id,
+        node_id=self.node_id,
+        node_data_memory=node_data.memory,
+        model_instance=model_instance,
+    )
+elif node_data.memory and node_data.memory.mode == MemoryMode.CONVERSATION:
+    # Conversation-level memory (existing behavior)
+    memory = fetch_memory(
+        variable_pool=variable_pool,
+        app_id=app_id,
+        node_data_memory=node_data.memory,
+        model_instance=model_instance,
+    )
+else:
+    memory = None
+
+# 2. Get history for context
+if memory:
+    if isinstance(memory, NodeTokenBufferMemory):
+        history = memory.get_history_prompt_messages(
+            current_message_id=current_message_id,
+            tenant_id=tenant_id,
+            max_token_limit=max_token_limit,
+        )
+    else:  # TokenBufferMemory
+        history = memory.get_history_prompt_messages(
+            max_token_limit=max_token_limit,
+        )
+    prompt_messages = [*history, *current_messages]
+else:
+    prompt_messages = current_messages
+
+# 3. Call LLM
+response = model_instance.invoke(prompt_messages)
+
+# 4. Append to node memory (only for NodeTokenBufferMemory)
+if isinstance(memory, NodeTokenBufferMemory):
+    memory.add_messages(
+        message_id=message_id,
+        parent_message_id=parent_message_id,
+        user_content=user_input,
+        user_files=user_files,
+        assistant_content=response.content,
+        assistant_files=response_files,
+    )
+    memory.flush()
+```
+
+### Configuration
+
+Add to `MemoryConfig` in `core/workflow/nodes/llm/entities.py`:
+
+```python
+class MemoryMode(StrEnum):
+    CONVERSATION = "conversation"  # Use TokenBufferMemory (default, existing behavior)
+    NODE = "node"                  # Use NodeTokenBufferMemory (new, Chatflow only)
+
+class MemoryConfig(BaseModel):
+    # Existing fields
+    role_prefix: RolePrefix | None = None
+    window: MemoryWindowConfig | None = None
+    query_prompt_template: str | None = None
+
+    # Memory mode (new)
+    mode: MemoryMode = MemoryMode.CONVERSATION
+```
+
+**Mode Behavior:**
+
+| Mode           | Memory Class          | Scope                    | Availability  |
+| -------------- | --------------------- | ------------------------ | ------------- |
+| `conversation` | TokenBufferMemory     | Entire conversation      | All app modes |
+| `node`         | NodeTokenBufferMemory | Per-node in conversation | Chatflow only |
+
+> When `mode=node` is used in a non-Chatflow context (no conversation_id), it should
+> fall back to no memory or raise a configuration error.
+
+---
+
+## Comparison
+
+| Feature        | TokenBufferMemory        | NodeTokenBufferMemory     |
+| -------------- | ------------------------ | ------------------------- |
+| Scope          | Conversation             | Node within Conversation  |
+| Storage        | Database (Message table) | Object Storage (JSON)     |
+| Thread Support | Yes                      | Yes                       |
+| File Support   | Yes (via MessageFile)    | Yes (via file references) |
+| Token Limit    | Yes                      | Yes                       |
+| Use Case       | Standard chat apps       | Complex workflows         |
+
+---
+
+## Future Considerations
+
+1. **Cleanup Task**: Add a Celery task to clean up old node memory files
+2. **Concurrency**: Consider Redis lock for concurrent node executions
+3. **Compression**: Compress large memory files to reduce storage costs
+4. **Extension**: Other nodes (Agent, Tool) may also benefit from node-level memory
@@ -0,0 +1,15 @@
+from core.memory.base import BaseMemory
+from core.memory.node_token_buffer_memory import (
+    NodeMemoryData,
+    NodeMemoryFile,
+    NodeTokenBufferMemory,
+)
+from core.memory.token_buffer_memory import TokenBufferMemory
+
+__all__ = [
+    "BaseMemory",
+    "NodeMemoryData",
+    "NodeMemoryFile",
+    "NodeTokenBufferMemory",
+    "TokenBufferMemory",
+]
@@ -0,0 +1,83 @@
+"""
+Base memory interfaces and types.
+
+This module defines the common protocol for memory implementations.
+"""
+
+from abc import ABC, abstractmethod
+from collections.abc import Sequence
+
+from core.model_runtime.entities import ImagePromptMessageContent, PromptMessage
+
+
+class BaseMemory(ABC):
+    """
+    Abstract base class for memory implementations.
+
+    Provides a common interface for both conversation-level and node-level memory.
+    """
+
+    @abstractmethod
+    def get_history_prompt_messages(
+        self,
+        *,
+        max_token_limit: int = 2000,
+        message_limit: int | None = None,
+    ) -> Sequence[PromptMessage]:
+        """
+        Get history prompt messages.
+
+        :param max_token_limit: Maximum tokens for history
+        :param message_limit: Maximum number of messages
+        :return: Sequence of PromptMessage for LLM context
+        """
+        pass
+
+    def get_history_prompt_text(
+        self,
+        human_prefix: str = "Human",
+        ai_prefix: str = "Assistant",
+        max_token_limit: int = 2000,
+        message_limit: int | None = None,
+    ) -> str:
+        """
+        Get history prompt as formatted text.
+
+        :param human_prefix: Prefix for human messages
+        :param ai_prefix: Prefix for assistant messages
+        :param max_token_limit: Maximum tokens for history
+        :param message_limit: Maximum number of messages
+        :return: Formatted history text
+        """
+        from core.model_runtime.entities import (
+            PromptMessageRole,
+            TextPromptMessageContent,
+        )
+
+        prompt_messages = self.get_history_prompt_messages(
+            max_token_limit=max_token_limit,
+            message_limit=message_limit,
+        )
+
+        string_messages = []
+        for m in prompt_messages:
+            if m.role == PromptMessageRole.USER:
+                role = human_prefix
+            elif m.role == PromptMessageRole.ASSISTANT:
+                role = ai_prefix
+            else:
+                continue
+
+            if isinstance(m.content, list):
+                inner_msg = ""
+                for content in m.content:
+                    if isinstance(content, TextPromptMessageContent):
+                        inner_msg += f"{content.data}\n"
+                    elif isinstance(content, ImagePromptMessageContent):
+                        inner_msg += "[image]\n"
+                string_messages.append(f"{role}: {inner_msg.strip()}")
+            else:
+                message = f"{role}: {m.content}"
+                string_messages.append(message)
+
+        return "\n".join(string_messages)
@@ -0,0 +1,353 @@
+"""
+Node-level Token Buffer Memory for Chatflow.
+
+This module provides node-scoped memory within a conversation.
+Each LLM node in a workflow can maintain its own independent conversation history.
+
+Note: This is only available in Chatflow (advanced-chat mode) because it requires
+both conversation_id and node_id.
+
+Design:
+- Storage is indexed by workflow_run_id (each execution stores one turn)
+- Thread tracking leverages Message table's parent_message_id structure
+- On read: query Message table for current thread, then filter Node Memory by workflow_run_ids
+"""
+
+import logging
+from collections.abc import Sequence
+
+from pydantic import BaseModel
+from sqlalchemy import select
+
+from core.file import File, FileTransferMethod
+from core.memory.base import BaseMemory
+from core.model_manager import ModelInstance
+from core.model_runtime.entities import (
+    AssistantPromptMessage,
+    ImagePromptMessageContent,
+    PromptMessage,
+    TextPromptMessageContent,
+    UserPromptMessage,
+)
+from core.prompt.utils.extract_thread_messages import extract_thread_messages
+from extensions.ext_database import db
+from extensions.ext_storage import storage
+from models.model import Message
+
+logger = logging.getLogger(__name__)
+
+
+class NodeMemoryFile(BaseModel):
+    """File reference stored in node memory."""
+
+    type: str  # image, audio, video, document, custom
+    transfer_method: str  # local_file, remote_url, tool_file
+    upload_file_id: str | None = None
+    tool_file_id: str | None = None
+    url: str | None = None
+
+
+class NodeMemoryTurn(BaseModel):
+    """A single dialogue turn (user + assistant) in node memory."""
+
+    user_content: str = ""
+    user_files: list[NodeMemoryFile] = []
+    assistant_content: str = ""
+    assistant_files: list[NodeMemoryFile] = []
+
+
+class NodeMemoryData(BaseModel):
+    """Root data structure for node memory storage."""
+
+    version: int = 1
+    # Key: workflow_run_id, Value: dialogue turn
+    turns: dict[str, NodeMemoryTurn] = {}
+
+
+class NodeTokenBufferMemory(BaseMemory):
+    """
+    Node-level Token Buffer Memory.
+
+    Provides node-scoped memory within a conversation. Each LLM node can maintain
+    its own independent conversation history, stored in object storage.
+
+    Key design: Thread tracking is delegated to Message table's parent_message_id.
+    Storage is indexed by workflow_run_id for easy filtering.
+
+    Storage key format: node_memory/{app_id}/{conversation_id}/{node_id}.json
+    """
+
+    def __init__(
+        self,
+        app_id: str,
+        conversation_id: str,
+        node_id: str,
+        tenant_id: str,
+        model_instance: ModelInstance,
+    ):
+        """
+        Initialize node-level memory.
+
+        :param app_id: Application ID
+        :param conversation_id: Conversation ID
+        :param node_id: Node ID in the workflow
+        :param tenant_id: Tenant ID for file reconstruction
+        :param model_instance: Model instance for token counting
+        """
+        self.app_id = app_id
+        self.conversation_id = conversation_id
+        self.node_id = node_id
+        self.tenant_id = tenant_id
+        self.model_instance = model_instance
+        self._storage_key = f"node_memory/{app_id}/{conversation_id}/{node_id}.json"
+        self._data: NodeMemoryData | None = None
+        self._dirty = False
+
+    def _load(self) -> NodeMemoryData:
+        """Load data from object storage."""
+        if self._data is not None:
+            return self._data
+
+        try:
+            raw = storage.load_once(self._storage_key)
+            self._data = NodeMemoryData.model_validate_json(raw)
+        except Exception:
+            # File not found or parse error, start fresh
+            self._data = NodeMemoryData()
+
+        return self._data
+
+    def _save(self) -> None:
+        """Save data to object storage."""
+        if self._data is not None:
+            storage.save(self._storage_key, self._data.model_dump_json().encode("utf-8"))
+            self._dirty = False
+
+    def _file_to_memory_file(self, file: File) -> NodeMemoryFile:
+        """Convert File object to NodeMemoryFile reference."""
+        return NodeMemoryFile(
+            type=file.type.value if hasattr(file.type, "value") else str(file.type),
+            transfer_method=(
+                file.transfer_method.value if hasattr(file.transfer_method, "value") else str(file.transfer_method)
+            ),
+            upload_file_id=file.related_id if file.transfer_method == FileTransferMethod.LOCAL_FILE else None,
+            tool_file_id=file.related_id if file.transfer_method == FileTransferMethod.TOOL_FILE else None,
+            url=file.remote_url if file.transfer_method == FileTransferMethod.REMOTE_URL else None,
+        )
+
+    def _memory_file_to_mapping(self, memory_file: NodeMemoryFile) -> dict:
+        """Convert NodeMemoryFile to mapping for file_factory."""
+        mapping: dict = {
+            "type": memory_file.type,
+            "transfer_method": memory_file.transfer_method,
+        }
+        if memory_file.upload_file_id:
+            mapping["upload_file_id"] = memory_file.upload_file_id
+        if memory_file.tool_file_id:
+            mapping["tool_file_id"] = memory_file.tool_file_id
+        if memory_file.url:
+            mapping["url"] = memory_file.url
+        return mapping
+
+    def _rebuild_files(self, memory_files: list[NodeMemoryFile]) -> list[File]:
+        """Rebuild File objects from NodeMemoryFile references."""
+        if not memory_files:
+            return []
+
+        from factories import file_factory
+
+        files = []
+        for mf in memory_files:
+            try:
+                mapping = self._memory_file_to_mapping(mf)
+                file = file_factory.build_from_mapping(mapping=mapping, tenant_id=self.tenant_id)
+                files.append(file)
+            except Exception as e:
+                logger.warning("Failed to rebuild file from memory: %s", e)
+                continue
+        return files
+
+    def _build_prompt_message(
+        self,
+        role: str,
+        content: str,
+        files: list[File],
+        detail: ImagePromptMessageContent.DETAIL = ImagePromptMessageContent.DETAIL.HIGH,
+    ) -> PromptMessage:
+        """Build PromptMessage from content and files."""
+        from core.file import file_manager
+
+        if not files:
+            if role == "user":
+                return UserPromptMessage(content=content)
+            else:
+                return AssistantPromptMessage(content=content)
+
+        # Build multimodal content
+        prompt_contents: list = []
+        for file in files:
+            try:
+                prompt_content = file_manager.to_prompt_message_content(file, image_detail_config=detail)
+                prompt_contents.append(prompt_content)
+            except Exception as e:
+                logger.warning("Failed to convert file to prompt content: %s", e)
+                continue
+
+        prompt_contents.append(TextPromptMessageContent(data=content))
+
+        if role == "user":
+            return UserPromptMessage(content=prompt_contents)
+        else:
+            return AssistantPromptMessage(content=prompt_contents)
+
+    def _get_thread_workflow_run_ids(self) -> list[str]:
+        """
+        Get workflow_run_ids for the current thread by querying Message table.
+
+        Returns workflow_run_ids in chronological order (oldest first).
+        """
+        # Query messages for this conversation
+        stmt = (
+            select(Message).where(Message.conversation_id == self.conversation_id).order_by(Message.created_at.desc())
+        )
+        messages = db.session.scalars(stmt.limit(500)).all()
+
+        if not messages:
+            return []
+
+        # Extract thread messages using existing logic
+        thread_messages = extract_thread_messages(messages)
+
+        # For newly created message, its answer is temporarily empty, skip it
+        if thread_messages and not thread_messages[0].answer and thread_messages[0].answer_tokens == 0:
+            thread_messages.pop(0)
+
+        # Reverse to get chronological order, extract workflow_run_ids
+        workflow_run_ids = []
+        for msg in reversed(thread_messages):
+            if msg.workflow_run_id:
+                workflow_run_ids.append(msg.workflow_run_id)
+
+        return workflow_run_ids
+
+    def add_messages(
+        self,
+        workflow_run_id: str,
+        user_content: str,
+        user_files: Sequence[File] | None = None,
+        assistant_content: str = "",
+        assistant_files: Sequence[File] | None = None,
+    ) -> None:
+        """
+        Add a dialogue turn to node memory.
+        Call this after LLM node execution completes.
+
+        :param workflow_run_id: Current workflow execution ID
+        :param user_content: User's text input
+        :param user_files: Files attached by user
+        :param assistant_content: Assistant's text response
+        :param assistant_files: Files generated by assistant
+        """
+        data = self._load()
+
+        # Convert files to memory file references
+        user_memory_files = [self._file_to_memory_file(f) for f in (user_files or [])]
+        assistant_memory_files = [self._file_to_memory_file(f) for f in (assistant_files or [])]
+
+        # Store the turn indexed by workflow_run_id
+        data.turns[workflow_run_id] = NodeMemoryTurn(
+            user_content=user_content,
+            user_files=user_memory_files,
+            assistant_content=assistant_content,
+            assistant_files=assistant_memory_files,
+        )
+
+        self._dirty = True
+
+    def get_history_prompt_messages(
+        self,
+        *,
+        max_token_limit: int = 2000,
+        message_limit: int | None = None,
+    ) -> Sequence[PromptMessage]:
+        """
+        Retrieve history as PromptMessage sequence.
+
+        Thread tracking is handled by querying Message table's parent_message_id structure.
+
+        :param max_token_limit: Maximum tokens for history
+        :param message_limit: unused, for interface compatibility
+        :return: Sequence of PromptMessage for LLM context
+        """
+        # message_limit is unused in NodeTokenBufferMemory (uses token limit instead)
+        _ = message_limit
+        detail = ImagePromptMessageContent.DETAIL.HIGH
+        data = self._load()
+
+        if not data.turns:
+            return []
+
+        # Get workflow_run_ids for current thread from Message table
+        thread_workflow_run_ids = self._get_thread_workflow_run_ids()
+
+        if not thread_workflow_run_ids:
+            return []
+
+        # Build prompt messages in thread order
+        prompt_messages: list[PromptMessage] = []
+        for wf_run_id in thread_workflow_run_ids:
+            turn = data.turns.get(wf_run_id)
+            if not turn:
+                # This workflow execution didn't have node memory stored
+                continue
+
+            # Build user message
+            user_files = self._rebuild_files(turn.user_files) if turn.user_files else []
+            user_msg = self._build_prompt_message(
+                role="user",
+                content=turn.user_content,
+                files=user_files,
+                detail=detail,
+            )
+            prompt_messages.append(user_msg)
+
+            # Build assistant message
+            assistant_files = self._rebuild_files(turn.assistant_files) if turn.assistant_files else []
+            assistant_msg = self._build_prompt_message(
+                role="assistant",
+                content=turn.assistant_content,
+                files=assistant_files,
+                detail=detail,
+            )
+            prompt_messages.append(assistant_msg)
+
+        if not prompt_messages:
+            return []
+
+        # Truncate by token limit
+        try:
+            current_tokens = self.model_instance.get_llm_num_tokens(prompt_messages)
+            while current_tokens > max_token_limit and len(prompt_messages) > 1:
+                prompt_messages.pop(0)
+                current_tokens = self.model_instance.get_llm_num_tokens(prompt_messages)
+        except Exception as e:
+            logger.warning("Failed to count tokens for truncation: %s", e)
+
+        return prompt_messages
+
+    def flush(self) -> None:
+        """
+        Persist buffered changes to object storage.
+        Call this at the end of node execution.
+        """
+        if self._dirty:
+            self._save()
+
+    def clear(self) -> None:
+        """Clear all messages in this node's memory."""
+        self._data = NodeMemoryData()
+        self._save()
+
+    def exists(self) -> bool:
+        """Check if node memory exists in storage."""
+        return storage.exists(self._storage_key)
@@ -5,12 +5,12 @@ from sqlalchemy.orm import sessionmaker

 from core.app.app_config.features.file_upload.manager import FileUploadConfigManager
 from core.file import file_manager
+from core.memory.base import BaseMemory
 from core.model_manager import ModelInstance
 from core.model_runtime.entities import (
    AssistantPromptMessage,
    ImagePromptMessageContent,
    PromptMessage,
-    PromptMessageRole,
    TextPromptMessageContent,
    UserPromptMessage,
 )
@@ -24,7 +24,7 @@ from repositories.api_workflow_run_repository import APIWorkflowRunRepository
 from repositories.factory import DifyAPIRepositoryFactory


-class TokenBufferMemory:
+class TokenBufferMemory(BaseMemory):
    def __init__(
        self,
        conversation: Conversation,
@@ -115,10 +115,14 @@ class TokenBufferMemory:
                return AssistantPromptMessage(content=prompt_message_contents)

    def get_history_prompt_messages(
-        self, max_token_limit: int = 2000, message_limit: int | None = None
+        self,
+        *,
+        max_token_limit: int = 2000,
+        message_limit: int | None = None,
    ) -> Sequence[PromptMessage]:
        """
        Get history prompt messages.
+
        :param max_token_limit: max token limit
        :param message_limit: message limit
        """
@@ -200,44 +204,3 @@ class TokenBufferMemory:
                curr_message_tokens = self.model_instance.get_llm_num_tokens(prompt_messages)

        return prompt_messages
-
-    def get_history_prompt_text(
-        self,
-        human_prefix: str = "Human",
-        ai_prefix: str = "Assistant",
-        max_token_limit: int = 2000,
-        message_limit: int | None = None,
-    ) -> str:
-        """
-        Get history prompt text.
-        :param human_prefix: human prefix
-        :param ai_prefix: ai prefix
-        :param max_token_limit: max token limit
-        :param message_limit: message limit
-        :return:
-        """
-        prompt_messages = self.get_history_prompt_messages(max_token_limit=max_token_limit, message_limit=message_limit)
-
-        string_messages = []
-        for m in prompt_messages:
-            if m.role == PromptMessageRole.USER:
-                role = human_prefix
-            elif m.role == PromptMessageRole.ASSISTANT:
-                role = ai_prefix
-            else:
-                continue
-
-            if isinstance(m.content, list):
-                inner_msg = ""
-                for content in m.content:
-                    if isinstance(content, TextPromptMessageContent):
-                        inner_msg += f"{content.data}\n"
-                    elif isinstance(content, ImagePromptMessageContent):
-                        inner_msg += "[image]\n"
-
-                string_messages.append(f"{role}: {inner_msg.strip()}")
-            else:
-                message = f"{role}: {m.content}"
-                string_messages.append(message)
-
-        return "\n".join(string_messages)
@@ -251,10 +251,7 @@ class AssistantPromptMessage(PromptMessage):

        :return: True if prompt message is empty, False otherwise
        """
-        if not super().is_empty() and not self.tool_calls:
-            return False
-
-        return True
+        return super().is_empty() and not self.tool_calls


 class SystemPromptMessage(PromptMessage):
@@ -1,6 +1,7 @@
 import logging
 from collections.abc import Sequence

+from opentelemetry.trace import SpanKind
 from sqlalchemy.orm import sessionmaker

 from core.ops.aliyun_trace.data_exporter.traceclient import (
@@ -54,7 +55,7 @@ from core.ops.entities.trace_entity import (
    ToolTraceInfo,
    WorkflowTraceInfo,
 )
-from core.repositories import SQLAlchemyWorkflowNodeExecutionRepository
+from core.repositories import DifyCoreRepositoryFactory
 from core.workflow.entities import WorkflowNodeExecution
 from core.workflow.enums import NodeType, WorkflowNodeExecutionMetadataKey
 from extensions.ext_database import db
@@ -151,6 +152,7 @@ class AliyunDataTrace(BaseTraceInstance):
            ),
            status=status,
            links=trace_metadata.links,
+            span_kind=SpanKind.SERVER,
        )
        self.trace_client.add_span(message_span)

@@ -273,7 +275,7 @@ class AliyunDataTrace(BaseTraceInstance):
        service_account = self.get_service_account_with_tenant(app_id)

        session_factory = sessionmaker(bind=db.engine)
-        workflow_node_execution_repository = SQLAlchemyWorkflowNodeExecutionRepository(
+        workflow_node_execution_repository = DifyCoreRepositoryFactory.create_workflow_node_execution_repository(
            session_factory=session_factory,
            user=service_account,
            app_id=app_id,
@@ -456,6 +458,7 @@ class AliyunDataTrace(BaseTraceInstance):
                ),
                status=status,
                links=trace_metadata.links,
+                span_kind=SpanKind.SERVER,
            )
            self.trace_client.add_span(message_span)

@@ -475,6 +478,7 @@ class AliyunDataTrace(BaseTraceInstance):
            ),
            status=status,
            links=trace_metadata.links,
+            span_kind=SpanKind.SERVER if message_span_id is None else SpanKind.INTERNAL,
        )
        self.trace_client.add_span(workflow_span)

@@ -166,7 +166,7 @@ class SpanBuilder:
            attributes=span_data.attributes,
            events=span_data.events,
            links=span_data.links,
-            kind=trace_api.SpanKind.INTERNAL,
+            kind=span_data.span_kind,
            status=span_data.status,
            start_time=span_data.start_time,
            end_time=span_data.end_time,
@@ -4,7 +4,7 @@ from typing import Any

 from opentelemetry import trace as trace_api
 from opentelemetry.sdk.trace import Event
-from opentelemetry.trace import Status, StatusCode
+from opentelemetry.trace import SpanKind, Status, StatusCode
 from pydantic import BaseModel, Field


@@ -34,3 +34,4 @@ class SpanData(BaseModel):
    status: Status = Field(default=Status(StatusCode.UNSET), description="The status of the span.")
    start_time: int | None = Field(..., description="The start time of the span in nanoseconds.")
    end_time: int | None = Field(..., description="The end time of the span in nanoseconds.")
+    span_kind: SpanKind = Field(default=SpanKind.INTERNAL, description="The OpenTelemetry SpanKind for this span.")
@@ -1,5 +1,6 @@
 from core.plugin.entities.endpoint import EndpointEntityWithInstance
 from core.plugin.impl.base import BasePluginClient
+from core.plugin.impl.exc import PluginDaemonInternalServerError


 class PluginEndpointClient(BasePluginClient):
@@ -70,18 +71,27 @@ class PluginEndpointClient(BasePluginClient):
    def delete_endpoint(self, tenant_id: str, user_id: str, endpoint_id: str):
        """
        Delete the given endpoint.
+
+        This operation is idempotent: if the endpoint is already deleted (record not found),
+        it will return True instead of raising an error.
        """
-        return self._request_with_plugin_daemon_response(
-            "POST",
-            f"plugin/{tenant_id}/endpoint/remove",
-            bool,
-            data={
-                "endpoint_id": endpoint_id,
-            },
-            headers={
-                "Content-Type": "application/json",
-            },
-        )
+        try:
+            return self._request_with_plugin_daemon_response(
+                "POST",
+                f"plugin/{tenant_id}/endpoint/remove",
+                bool,
+                data={
+                    "endpoint_id": endpoint_id,
+                },
+                headers={
+                    "Content-Type": "application/json",
+                },
+            )
+        except PluginDaemonInternalServerError as e:
+            # Make delete idempotent: if record is not found, consider it a success
+            if "record not found" in str(e.description).lower():
+                return True
+            raise

    def enable_endpoint(self, tenant_id: str, user_id: str, endpoint_id: str):
        """
@@ -5,7 +5,7 @@ from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEnti
 from core.file import file_manager
 from core.file.models import File
 from core.helper.code_executor.jinja2.jinja2_formatter import Jinja2Formatter
-from core.memory.token_buffer_memory import TokenBufferMemory
+from core.memory.base import BaseMemory
 from core.model_runtime.entities import (
    AssistantPromptMessage,
    PromptMessage,
@@ -43,7 +43,7 @@ class AdvancedPromptTransform(PromptTransform):
        files: Sequence[File],
        context: str | None,
        memory_config: MemoryConfig | None,
-        memory: TokenBufferMemory | None,
+        memory: BaseMemory | None,
        model_config: ModelConfigWithCredentialsEntity,
        image_detail_config: ImagePromptMessageContent.DETAIL | None = None,
    ) -> list[PromptMessage]:
@@ -84,7 +84,7 @@ class AdvancedPromptTransform(PromptTransform):
        files: Sequence[File],
        context: str | None,
        memory_config: MemoryConfig | None,
-        memory: TokenBufferMemory | None,
+        memory: BaseMemory | None,
        model_config: ModelConfigWithCredentialsEntity,
        image_detail_config: ImagePromptMessageContent.DETAIL | None = None,
    ) -> list[PromptMessage]:
@@ -145,7 +145,7 @@ class AdvancedPromptTransform(PromptTransform):
        files: Sequence[File],
        context: str | None,
        memory_config: MemoryConfig | None,
-        memory: TokenBufferMemory | None,
+        memory: BaseMemory | None,
        model_config: ModelConfigWithCredentialsEntity,
        image_detail_config: ImagePromptMessageContent.DETAIL | None = None,
    ) -> list[PromptMessage]:
@@ -270,7 +270,7 @@ class AdvancedPromptTransform(PromptTransform):

    def _set_histories_variable(
        self,
-        memory: TokenBufferMemory,
+        memory: BaseMemory,
        memory_config: MemoryConfig,
        raw_prompt: str,
        role_prefix: MemoryConfig.RolePrefix,
@@ -1,3 +1,4 @@
+from enum import StrEnum
 from typing import Literal

 from pydantic import BaseModel
@@ -5,6 +6,13 @@ from pydantic import BaseModel
 from core.model_runtime.entities.message_entities import PromptMessageRole


+class MemoryMode(StrEnum):
+    """Memory mode for LLM nodes."""
+
+    CONVERSATION = "conversation"  # Use TokenBufferMemory (default, existing behavior)
+    NODE = "node"  # Use NodeTokenBufferMemory (Chatflow only)
+
+
 class ChatModelMessage(BaseModel):
    """
    Chat Message.
@@ -48,3 +56,4 @@ class MemoryConfig(BaseModel):
    role_prefix: RolePrefix | None = None
    window: WindowConfig
    query_prompt_template: str | None = None
+    mode: MemoryMode = MemoryMode.CONVERSATION
@@ -1,7 +1,7 @@
 from typing import Any

 from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEntity
-from core.memory.token_buffer_memory import TokenBufferMemory
+from core.memory.base import BaseMemory
 from core.model_manager import ModelInstance
 from core.model_runtime.entities.message_entities import PromptMessage
 from core.model_runtime.entities.model_entities import ModelPropertyKey
@@ -11,7 +11,7 @@ from core.prompt.entities.advanced_prompt_entities import MemoryConfig
 class PromptTransform:
    def _append_chat_histories(
        self,
-        memory: TokenBufferMemory,
+        memory: BaseMemory,
        memory_config: MemoryConfig,
        prompt_messages: list[PromptMessage],
        model_config: ModelConfigWithCredentialsEntity,
@@ -52,7 +52,7 @@ class PromptTransform:

    def _get_history_messages_from_memory(
        self,
-        memory: TokenBufferMemory,
+        memory: BaseMemory,
        memory_config: MemoryConfig,
        max_token_limit: int,
        human_prefix: str | None = None,
@@ -73,7 +73,7 @@ class PromptTransform:
        return memory.get_history_prompt_text(**kwargs)

    def _get_history_messages_list_from_memory(
-        self, memory: TokenBufferMemory, memory_config: MemoryConfig, max_token_limit: int
+        self, memory: BaseMemory, memory_config: MemoryConfig, max_token_limit: int
    ) -> list[PromptMessage]:
        """Get memory messages."""
        return list(
@@ -618,18 +618,18 @@ class ProviderManager:
                )

            for quota in configuration.quotas:
-                if quota.quota_type == ProviderQuotaType.TRIAL:
+                if quota.quota_type in (ProviderQuotaType.TRIAL, ProviderQuotaType.PAID):
                    # Init trial provider records if not exists
-                    if ProviderQuotaType.TRIAL not in provider_quota_to_provider_record_dict:
+                    if quota.quota_type not in provider_quota_to_provider_record_dict:
                        try:
                            # FIXME ignore the type error, only TrialHostingQuota has limit need to change the logic
                            new_provider_record = Provider(
                                tenant_id=tenant_id,
                                # TODO: Use provider name with prefix after the data migration.
                                provider_name=ModelProviderID(provider_name).provider_name,
-                                provider_type=ProviderType.SYSTEM,
-                                quota_type=ProviderQuotaType.TRIAL,
-                                quota_limit=quota.quota_limit,  # type: ignore
+                                provider_type=ProviderType.SYSTEM.value,
+                                quota_type=quota.quota_type,
+                                quota_limit=0,  # type: ignore
                                quota_used=0,
                                is_valid=True,
                            )
@@ -641,8 +641,8 @@ class ProviderManager:
                            stmt = select(Provider).where(
                                Provider.tenant_id == tenant_id,
                                Provider.provider_name == ModelProviderID(provider_name).provider_name,
-                                Provider.provider_type == ProviderType.SYSTEM,
-                                Provider.quota_type == ProviderQuotaType.TRIAL,
+                                Provider.provider_type == ProviderType.SYSTEM.value,
+                                Provider.quota_type == quota.quota_type,
                            )
                            existed_provider_record = db.session.scalar(stmt)
                            if not existed_provider_record:
@@ -912,6 +912,22 @@ class ProviderManager:
                provider_record
            )
        quota_configurations = []
+
+        if dify_config.EDITION == "CLOUD":
+            from services.credit_pool_service import CreditPoolService
+
+            trail_pool = CreditPoolService.get_pool(
+                tenant_id=tenant_id,
+                pool_type=ProviderQuotaType.TRIAL.value,
+            )
+            paid_pool = CreditPoolService.get_pool(
+                tenant_id=tenant_id,
+                pool_type=ProviderQuotaType.PAID.value,
+            )
+        else:
+            trail_pool = None
+            paid_pool = None
+
        for provider_quota in provider_hosting_configuration.quotas:
            if provider_quota.quota_type not in quota_type_to_provider_records_dict:
                if provider_quota.quota_type == ProviderQuotaType.FREE:
@@ -932,16 +948,36 @@ class ProviderManager:
                    raise ValueError("quota_used is None")
                if provider_record.quota_limit is None:
                    raise ValueError("quota_limit is None")
+                if provider_quota.quota_type == ProviderQuotaType.TRIAL and trail_pool is not None:
+                    quota_configuration = QuotaConfiguration(
+                        quota_type=provider_quota.quota_type,
+                        quota_unit=provider_hosting_configuration.quota_unit or QuotaUnit.TOKENS,
+                        quota_used=trail_pool.quota_used,
+                        quota_limit=trail_pool.quota_limit,
+                        is_valid=trail_pool.quota_limit > trail_pool.quota_used or trail_pool.quota_limit == -1,
+                        restrict_models=provider_quota.restrict_models,
+                    )

-                quota_configuration = QuotaConfiguration(
-                    quota_type=provider_quota.quota_type,
-                    quota_unit=provider_hosting_configuration.quota_unit or QuotaUnit.TOKENS,
-                    quota_used=provider_record.quota_used,
-                    quota_limit=provider_record.quota_limit,
-                    is_valid=provider_record.quota_limit > provider_record.quota_used
-                    or provider_record.quota_limit == -1,
-                    restrict_models=provider_quota.restrict_models,
-                )
+                elif provider_quota.quota_type == ProviderQuotaType.PAID and paid_pool is not None:
+                    quota_configuration = QuotaConfiguration(
+                        quota_type=provider_quota.quota_type,
+                        quota_unit=provider_hosting_configuration.quota_unit or QuotaUnit.TOKENS,
+                        quota_used=paid_pool.quota_used,
+                        quota_limit=paid_pool.quota_limit,
+                        is_valid=paid_pool.quota_limit > paid_pool.quota_used or paid_pool.quota_limit == -1,
+                        restrict_models=provider_quota.restrict_models,
+                    )
+
+                else:
+                    quota_configuration = QuotaConfiguration(
+                        quota_type=provider_quota.quota_type,
+                        quota_unit=provider_hosting_configuration.quota_unit or QuotaUnit.TOKENS,
+                        quota_used=provider_record.quota_used,
+                        quota_limit=provider_record.quota_limit,
+                        is_valid=provider_record.quota_limit > provider_record.quota_used
+                        or provider_record.quota_limit == -1,
+                        restrict_models=provider_quota.restrict_models,
+                    )

            quota_configurations.append(quota_configuration)

@@ -29,7 +29,6 @@ from models import (
    Account,
    CreatorUserRole,
    EndUser,
-    LLMGenerationDetail,
    WorkflowNodeExecutionModel,
    WorkflowNodeExecutionTriggeredFrom,
 )
@@ -458,113 +457,6 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
            session.merge(db_model)
            session.flush()

-            # Save LLMGenerationDetail for LLM nodes with successful execution
-            if (
-                domain_model.node_type == NodeType.LLM
-                and domain_model.status == WorkflowNodeExecutionStatus.SUCCEEDED
-                and domain_model.outputs is not None
-            ):
-                self._save_llm_generation_detail(session, domain_model)
-
-    def _save_llm_generation_detail(self, session, execution: WorkflowNodeExecution) -> None:
-        """
-        Save LLM generation detail for LLM nodes.
-        Extracts reasoning_content, tool_calls, and sequence from outputs and metadata.
-        """
-        outputs = execution.outputs or {}
-        metadata = execution.metadata or {}
-
-        reasoning_list = self._extract_reasoning(outputs)
-        tool_calls_list = self._extract_tool_calls(metadata.get(WorkflowNodeExecutionMetadataKey.AGENT_LOG))
-
-        if not reasoning_list and not tool_calls_list:
-            return
-
-        sequence = self._build_generation_sequence(outputs.get("text", ""), reasoning_list, tool_calls_list)
-        self._upsert_generation_detail(session, execution, reasoning_list, tool_calls_list, sequence)
-
-    def _extract_reasoning(self, outputs: Mapping[str, Any]) -> list[str]:
-        """Extract reasoning_content as a clean list of non-empty strings."""
-        reasoning_content = outputs.get("reasoning_content")
-        if isinstance(reasoning_content, str):
-            trimmed = reasoning_content.strip()
-            return [trimmed] if trimmed else []
-        if isinstance(reasoning_content, list):
-            return [item.strip() for item in reasoning_content if isinstance(item, str) and item.strip()]
-        return []
-
-    def _extract_tool_calls(self, agent_log: Any) -> list[dict[str, str]]:
-        """Extract tool call records from agent logs."""
-        if not agent_log or not isinstance(agent_log, list):
-            return []
-
-        tool_calls: list[dict[str, str]] = []
-        for log in agent_log:
-            log_data = log.data if hasattr(log, "data") else (log.get("data", {}) if isinstance(log, dict) else {})
-            tool_name = log_data.get("tool_name")
-            if tool_name and str(tool_name).strip():
-                tool_calls.append(
-                    {
-                        "id": log_data.get("tool_call_id", ""),
-                        "name": tool_name,
-                        "arguments": json.dumps(log_data.get("tool_args", {})),
-                        "result": str(log_data.get("output", "")),
-                    }
-                )
-        return tool_calls
-
-    def _build_generation_sequence(
-        self, text: str, reasoning_list: list[str], tool_calls_list: list[dict[str, str]]
-    ) -> list[dict[str, Any]]:
-        """Build a simple content/reasoning/tool_call sequence."""
-        sequence: list[dict[str, Any]] = []
-        if text:
-            sequence.append({"type": "content", "start": 0, "end": len(text)})
-        for index in range(len(reasoning_list)):
-            sequence.append({"type": "reasoning", "index": index})
-        for index in range(len(tool_calls_list)):
-            sequence.append({"type": "tool_call", "index": index})
-        return sequence
-
-    def _upsert_generation_detail(
-        self,
-        session,
-        execution: WorkflowNodeExecution,
-        reasoning_list: list[str],
-        tool_calls_list: list[dict[str, str]],
-        sequence: list[dict[str, Any]],
-    ) -> None:
-        """Insert or update LLMGenerationDetail with serialized fields."""
-        existing = (
-            session.query(LLMGenerationDetail)
-            .filter_by(
-                workflow_run_id=execution.workflow_execution_id,
-                node_id=execution.node_id,
-            )
-            .first()
-        )
-
-        reasoning_json = json.dumps(reasoning_list) if reasoning_list else None
-        tool_calls_json = json.dumps(tool_calls_list) if tool_calls_list else None
-        sequence_json = json.dumps(sequence) if sequence else None
-
-        if existing:
-            existing.reasoning_content = reasoning_json
-            existing.tool_calls = tool_calls_json
-            existing.sequence = sequence_json
-            return
-
-        generation_detail = LLMGenerationDetail(
-            tenant_id=self._tenant_id,
-            app_id=self._app_id,
-            workflow_run_id=execution.workflow_execution_id,
-            node_id=execution.node_id,
-            reasoning_content=reasoning_json,
-            tool_calls=tool_calls_json,
-            sequence=sequence_json,
-        )
-        session.add(generation_detail)
-
    def get_db_models_by_workflow_run(
        self,
        workflow_run_id: str,
@@ -8,7 +8,6 @@ from typing import TYPE_CHECKING, Any
 if TYPE_CHECKING:
    from models.model import File

-from core.model_runtime.entities.message_entities import PromptMessageTool
 from core.tools.__base.tool_runtime import ToolRuntime
 from core.tools.entities.tool_entities import (
    ToolEntity,
@@ -155,60 +154,6 @@ class Tool(ABC):

        return parameters

-    def to_prompt_message_tool(self) -> PromptMessageTool:
-        message_tool = PromptMessageTool(
-            name=self.entity.identity.name,
-            description=self.entity.description.llm if self.entity.description else "",
-            parameters={
-                "type": "object",
-                "properties": {},
-                "required": [],
-            },
-        )
-
-        parameters = self.get_merged_runtime_parameters()
-        for parameter in parameters:
-            if parameter.form != ToolParameter.ToolParameterForm.LLM:
-                continue
-
-            parameter_type = parameter.type.as_normal_type()
-            if parameter.type in {
-                ToolParameter.ToolParameterType.SYSTEM_FILES,
-                ToolParameter.ToolParameterType.FILE,
-                ToolParameter.ToolParameterType.FILES,
-            }:
-                # Determine the description based on parameter type
-                if parameter.type == ToolParameter.ToolParameterType.FILE:
-                    file_format_desc = " Input the file id with format: [File: file_id]."
-                else:
-                    file_format_desc = "Input the file id with format: [Files: file_id1, file_id2, ...]. "
-
-                message_tool.parameters["properties"][parameter.name] = {
-                    "type": "string",
-                    "description": (parameter.llm_description or "") + file_format_desc,
-                }
-                continue
-            enum = []
-            if parameter.type == ToolParameter.ToolParameterType.SELECT:
-                enum = [option.value for option in parameter.options] if parameter.options else []
-
-            message_tool.parameters["properties"][parameter.name] = (
-                {
-                    "type": parameter_type,
-                    "description": parameter.llm_description or "",
-                }
-                if parameter.input_schema is None
-                else parameter.input_schema
-            )
-
-            if len(enum) > 0:
-                message_tool.parameters["properties"][parameter.name]["enum"] = enum
-
-            if parameter.required:
-                message_tool.parameters["required"].append(parameter.name)
-
-        return message_tool
-
    def create_image_message(
        self,
        image: str,
@@ -1047,6 +1047,8 @@ class ToolManager:
                        continue
                    tool_input = ToolNodeData.ToolInput.model_validate(tool_configurations.get(parameter.name, {}))
                    if tool_input.type == "variable":
+                        if not isinstance(tool_input.value, list):
+                            raise ToolParameterError(f"Invalid variable selector for {parameter.name}")
                        variable = variable_pool.get(tool_input.value)
                        if variable is None:
                            raise ToolParameterError(f"Variable {tool_input.value} does not exist")
@@ -1056,6 +1058,11 @@ class ToolManager:
                    elif tool_input.type == "mixed":
                        segment_group = variable_pool.convert_template(str(tool_input.value))
                        parameter_value = segment_group.text
+                    elif tool_input.type == "mention":
+                        # Mention type not supported in agent mode
+                        raise ToolParameterError(
+                            f"Mention type not supported in agent for parameter '{parameter.name}'"
+                        )
                    else:
                        raise ToolParameterError(f"Unknown tool input type '{tool_input.type}'")
                    runtime_parameters[parameter.name] = parameter_value
@@ -7,8 +7,8 @@ from typing import Any, cast

 from flask import has_request_context
 from sqlalchemy import select
-from sqlalchemy.orm import Session

+from core.db.session_factory import session_factory
 from core.file import FILE_MODEL_IDENTITY, File, FileTransferMethod
 from core.model_runtime.entities.llm_entities import LLMUsage, LLMUsageMetadata
 from core.tools.__base.tool import Tool
@@ -20,7 +20,6 @@ from core.tools.entities.tool_entities import (
    ToolProviderType,
 )
 from core.tools.errors import ToolInvokeError
-from extensions.ext_database import db
 from factories.file_factory import build_from_mapping
 from libs.login import current_user
 from models import Account, Tenant
@@ -230,30 +229,32 @@ class WorkflowTool(Tool):
        """
        Resolve user from database (worker/Celery context).
        """
+        with session_factory.create_session() as session:
+            tenant_stmt = select(Tenant).where(Tenant.id == self.runtime.tenant_id)
+            tenant = session.scalar(tenant_stmt)
+            if not tenant:
+                return None
+
+            user_stmt = select(Account).where(Account.id == user_id)
+            user = session.scalar(user_stmt)
+            if user:
+                user.current_tenant = tenant
+                session.expunge(user)
+                return user
+
+            end_user_stmt = select(EndUser).where(EndUser.id == user_id, EndUser.tenant_id == tenant.id)
+            end_user = session.scalar(end_user_stmt)
+            if end_user:
+                session.expunge(end_user)
+                return end_user

-        tenant_stmt = select(Tenant).where(Tenant.id == self.runtime.tenant_id)
-        tenant = db.session.scalar(tenant_stmt)
-        if not tenant:
            return None

-        user_stmt = select(Account).where(Account.id == user_id)
-        user = db.session.scalar(user_stmt)
-        if user:
-            user.current_tenant = tenant
-            return user
-
-        end_user_stmt = select(EndUser).where(EndUser.id == user_id, EndUser.tenant_id == tenant.id)
-        end_user = db.session.scalar(end_user_stmt)
-        if end_user:
-            return end_user
-
-        return None
-
    def _get_workflow(self, app_id: str, version: str) -> Workflow:
        """
        get the workflow by app id and version
        """
-        with Session(db.engine, expire_on_commit=False) as session, session.begin():
+        with session_factory.create_session() as session, session.begin():
            if not version:
                stmt = (
                    select(Workflow)
@@ -265,22 +266,24 @@ class WorkflowTool(Tool):
                stmt = select(Workflow).where(Workflow.app_id == app_id, Workflow.version == version)
                workflow = session.scalar(stmt)

-        if not workflow:
-            raise ValueError("workflow not found or not published")
+            if not workflow:
+                raise ValueError("workflow not found or not published")

-        return workflow
+            session.expunge(workflow)
+            return workflow

    def _get_app(self, app_id: str) -> App:
        """
        get the app by app id
        """
        stmt = select(App).where(App.id == app_id)
-        with Session(db.engine, expire_on_commit=False) as session, session.begin():
+        with session_factory.create_session() as session, session.begin():
            app = session.scalar(stmt)
-        if not app:
-            raise ValueError("app not found")
+            if not app:
+                raise ValueError("app not found")

-        return app
+            session.expunge(app)
+            return app

    def _transform_args(self, tool_parameters: dict) -> tuple[dict, list[dict]]:
        """
@@ -4,6 +4,7 @@ from .segments import (
    ArrayFileSegment,
    ArrayNumberSegment,
    ArrayObjectSegment,
+    ArrayPromptMessageSegment,
    ArraySegment,
    ArrayStringSegment,
    FileSegment,
@@ -20,6 +21,7 @@ from .variables import (
    ArrayFileVariable,
    ArrayNumberVariable,
    ArrayObjectVariable,
+    ArrayPromptMessageVariable,
    ArrayStringVariable,
    ArrayVariable,
    FileVariable,
@@ -30,6 +32,7 @@ from .variables import (
    SecretVariable,
    StringVariable,
    Variable,
+    VariableBase,
 )

 __all__ = [
@@ -41,6 +44,8 @@ __all__ = [
    "ArrayNumberVariable",
    "ArrayObjectSegment",
    "ArrayObjectVariable",
+    "ArrayPromptMessageSegment",
+    "ArrayPromptMessageVariable",
    "ArraySegment",
    "ArrayStringSegment",
    "ArrayStringVariable",
@@ -62,4 +67,5 @@ __all__ = [
    "StringSegment",
    "StringVariable",
    "Variable",
+    "VariableBase",
 ]
@@ -6,6 +6,7 @@ from typing import Annotated, Any, TypeAlias
 from pydantic import BaseModel, ConfigDict, Discriminator, Tag, field_validator

 from core.file import File
+from core.model_runtime.entities import PromptMessage

 from .types import SegmentType

@@ -208,6 +209,15 @@ class ArrayBooleanSegment(ArraySegment):
    value: Sequence[bool]


+class ArrayPromptMessageSegment(ArraySegment):
+    value_type: SegmentType = SegmentType.ARRAY_PROMPT_MESSAGE
+    value: Sequence[PromptMessage]
+
+    def to_object(self):
+        """Convert to JSON-serializable format for database storage and frontend."""
+        return [msg.model_dump() for msg in self.value]
+
+
 def get_segment_discriminator(v: Any) -> SegmentType | None:
    if isinstance(v, Segment):
        return v.value_type
@@ -232,7 +242,7 @@ def get_segment_discriminator(v: Any) -> SegmentType | None:
 # - All variants in `SegmentUnion` must inherit from the `Segment` class.
 # - The union must include all non-abstract subclasses of `Segment`, except:
 #   - `SegmentGroup`, which is not added to the variable pool.
-#   - `Variable` and its subclasses, which are handled by `VariableUnion`.
+#   - `VariableBase` and its subclasses, which are handled by `Variable`.
 SegmentUnion: TypeAlias = Annotated[
    (
        Annotated[NoneSegment, Tag(SegmentType.NONE)]
@@ -248,6 +258,7 @@ SegmentUnion: TypeAlias = Annotated[
        | Annotated[ArrayObjectSegment, Tag(SegmentType.ARRAY_OBJECT)]
        | Annotated[ArrayFileSegment, Tag(SegmentType.ARRAY_FILE)]
        | Annotated[ArrayBooleanSegment, Tag(SegmentType.ARRAY_BOOLEAN)]
+        | Annotated[ArrayPromptMessageSegment, Tag(SegmentType.ARRAY_PROMPT_MESSAGE)]
    ),
    Discriminator(get_segment_discriminator),
 ]
@@ -45,6 +45,7 @@ class SegmentType(StrEnum):
    ARRAY_OBJECT = "array[object]"
    ARRAY_FILE = "array[file]"
    ARRAY_BOOLEAN = "array[boolean]"
+    ARRAY_PROMPT_MESSAGE = "array[message]"

    NONE = "none"

--- a/Show More
+++ b/Show More