chore: Bump version

Signed-off-by: -LAN- <[email protected]>
fix: Refactor web reader to use readabilipy (#19789 )
2025-05-30 16:25:40 +08:00 · 2025-05-30 16:23:17 +08:00 · 2025-05-15 14:47:15 +08:00 · 2025-05-14 18:03:05 +08:00 · 2025-05-14 18:02:32 +08:00 · 2025-05-14 14:08:05 +08:00
240 changed files with 10193 additions and 9953 deletions
@@ -26,9 +26,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Setup Poetry and Python ${{ matrix.python-version }}
        uses: ./.github/actions/setup-poetry
@@ -5,8 +5,8 @@ on:
    branches:
      - "main"
      - "deploy/dev"
-  release:
-    types: [published]
+    tags:
+      - "*"

 concurrency:
  group: build-push-${{ github.head_ref || github.run_id }}
@@ -79,12 +79,10 @@ jobs:
          cache-to: type=gha,mode=max,scope=${{ matrix.service_name }}

      - name: Export digest
-        env:
-          DIGEST: ${{ steps.build.outputs.digest }}
        run: |
          mkdir -p /tmp/digests
-          sanitized_digest=${DIGEST#sha256:}
-          touch "/tmp/digests/${sanitized_digest}"
+          digest="${{ steps.build.outputs.digest }}"
+          touch "/tmp/digests/${digest#sha256:}"

      - name: Upload digest
        uses: actions/upload-artifact@v4
@@ -134,15 +132,10 @@ jobs:

      - name: Create manifest list and push
        working-directory: /tmp/digests
-        env:
-          IMAGE_NAME: ${{ env[matrix.image_name_env] }}
        run: |
          docker buildx imagetools create $(jq -cr '.tags | map("-t " + .) | join(" ")' <<< "$DOCKER_METADATA_OUTPUT_JSON") \
-            $(printf "$IMAGE_NAME@sha256:%s " *)
+            $(printf '${{ env[matrix.image_name_env] }}@sha256:%s ' *)

      - name: Inspect image
-        env:
-          IMAGE_NAME: ${{ env[matrix.image_name_env] }}
-          IMAGE_VERSION: ${{ steps.meta.outputs.version }}
        run: |
-          docker buildx imagetools inspect "$IMAGE_NAME:$IMAGE_VERSION"
+          docker buildx imagetools inspect ${{ env[matrix.image_name_env] }}:${{ steps.meta.outputs.version }}
@@ -19,9 +19,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Setup Poetry and Python
        uses: ./.github/actions/setup-poetry
@@ -9,6 +9,6 @@ yq eval '.services["pgvecto-rs"].ports += ["5431:5432"]' -i docker/docker-compos
 yq eval '.services["elasticsearch"].ports += ["9200:9200"]' -i docker/docker-compose.yaml
 yq eval '.services.couchbase-server.ports += ["8091-8096:8091-8096"]' -i docker/docker-compose.yaml
 yq eval '.services.couchbase-server.ports += ["11210:11210"]' -i docker/docker-compose.yaml
-yq eval '.services.tidb.ports += ["4000:4000"]' -i docker/tidb/docker-compose.yaml
+yq eval '.services.tidb.ports += ["4000:4000"]' -i docker/docker-compose.yaml

 echo "Ports exposed for sandbox, weaviate, tidb, qdrant, chroma, milvus, pgvector, pgvecto-rs, elasticsearch, couchbase"
@@ -17,9 +17,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Check changed files
        id: changed-files
@@ -62,9 +59,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Check changed files
        id: changed-files
@@ -95,9 +89,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Check changed files
        id: changed-files
@@ -126,9 +117,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Check changed files
        id: changed-files
@@ -26,9 +26,6 @@ jobs:

    steps:
      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Use Node.js ${{ matrix.node-version }}
        uses: actions/setup-node@v4
@@ -16,7 +16,6 @@ jobs:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 2 # last 2 commits
-          persist-credentials: false

      - name: Check for file changes in i18n/en-US
        id: check_files
@@ -28,9 +28,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Setup Poetry and Python ${{ matrix.python-version }}
        uses: ./.github/actions/setup-poetry
@@ -54,15 +51,7 @@ jobs:
      - name: Expose Service Ports
        run: sh .github/workflows/expose_service_ports.sh

-      - name: Set up Vector Store (TiDB)
-        uses: hoverkraft-tech/[email protected]
-        with:
-          compose-file: docker/tidb/docker-compose.yaml
-          services: |
-            tidb
-            tiflash
-
-      - name: Set up Vector Stores (Weaviate, Qdrant, PGVector, Milvus, PgVecto-RS, Chroma, MyScale, ElasticSearch, Couchbase)
+      - name: Set up Vector Stores (TiDB, Weaviate, Qdrant, PGVector, Milvus, PgVecto-RS, Chroma, MyScale, ElasticSearch, Couchbase)
        uses: hoverkraft-tech/[email protected]
        with:
          compose-file: |
@@ -78,9 +67,7 @@ jobs:
            pgvector
            chroma
            elasticsearch
-
-      - name: Check TiDB Ready
-        run: poetry run -P api python api/tests/integration_tests/vdb/tidb_vector/check_tiflash_ready.py
+            tidb

      - name: Test Vector Stores
        run: poetry run -P api bash dev/pytest/pytest_vdb.sh
@@ -22,9 +22,6 @@ jobs:
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          persist-credentials: false

      - name: Check changed files
        id: changed-files
@@ -163,7 +163,6 @@ docker/volumes/db/data/*
 docker/volumes/redis/data/*
 docker/volumes/weaviate/*
 docker/volumes/qdrant/*
-docker/tidb/volumes/*
 docker/volumes/etcd/*
 docker/volumes/minio/*
 docker/volumes/milvus/*
@@ -0,0 +1,4 @@
+{
+    "MD024": false,
+    "MD013": false
+}
@@ -0,0 +1,45 @@
+# Changelog
+
+All notable changes to Dify will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.15.8] - 2025-05-30
+
+### Added
+
+- Added gunicorn keepalive setting (#19537)
+
+### Fixed
+
+- Fixed database configuration to allow DB_EXTRAS to set search_path via options (#16a4f77)
+- Fixed frontend third-party package security issues (#19655)
+- Updated dependencies: huggingface-hub (~0.16.4 to ~0.31.0), transformers (~4.35.0 to ~4.39.0), and resend (~0.7.0 to ~2.9.0) (#19563)
+- Downgrade boto3 from 1.36 to 1.35 (#19736)
+
+## [0.15.7] - 2025-04-27
+
+### Added
+
+- Added support for GPT-4.1 in model providers (#18912)
+- Added support for Amazon Bedrock DeepSeek-R1 model (#18908)
+- Added support for Amazon Bedrock Claude Sonnet 3.7 model (#18788)
+- Refined version compatibility logic in app DSL service
+
+### Fixed
+
+- Fixed issue with creating apps from template categories (#18807, #18868)
+- Fixed DSL version check when creating apps from explore templates (#18872, #18878)
+
+## [0.15.6] - 2025-04-22
+
+### Security
+
+- Fixed clickjacking vulnerability (#18552)
+- Fixed reset password security issue (#18366)
+- Updated reset password token when email code verification succeeds (#18362)
+
+### Fixed
+
+- Fixed Vertex AI Gemini 2.0 Flash 001 schema (#18405)
@@ -108,72 +108,6 @@ Please refer to our [FAQ](https://docs.dify.ai/getting-started/install-self-host
 **7. Backend-as-a-Service**: 
  All of Dify's offerings come with corresponding APIs, so you could effortlessly integrate Dify into your own business logic.

-## Feature Comparison
-<table style="width: 100%;">
-  <tr>
-    <th align="center">Feature</th>
-    <th align="center">Dify.AI</th>
-    <th align="center">LangChain</th>
-    <th align="center">Flowise</th>
-    <th align="center">OpenAI Assistants API</th>
-  </tr>
-  <tr>
-    <td align="center">Programming Approach</td>
-    <td align="center">API + App-oriented</td>
-    <td align="center">Python Code</td>
-    <td align="center">App-oriented</td>
-    <td align="center">API-oriented</td>
-  </tr>
-  <tr>
-    <td align="center">Supported LLMs</td>
-    <td align="center">Rich Variety</td>
-    <td align="center">Rich Variety</td>
-    <td align="center">Rich Variety</td>
-    <td align="center">OpenAI-only</td>
-  </tr>
-  <tr>
-    <td align="center">RAG Engine</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-  </tr>
-  <tr>
-    <td align="center">Agent</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">✅</td>
-  </tr>
-  <tr>
-    <td align="center">Workflow</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Observability</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Enterprise Feature (SSO/Access control)</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Local Deployment</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-  </tr>
-</table>

 ## Using Dify

@@ -87,7 +87,9 @@ Dify is an open-source LLM app development platform. Its intuitive interface com

 ## Feature Comparison
 <table style="width: 100%;">
-  <tr>
+  <tr
+
+>
    <th align="center">Feature</th>
    <th align="center">Dify.AI</th>
    <th align="center">LangChain</th>
@@ -106,73 +106,6 @@ Prosimo, glejte naša pogosta vprašanja [FAQ](https://docs.dify.ai/getting-star
 **7. Backend-as-a-Service**: 
  AVse ponudbe Difyja so opremljene z ustreznimi API-ji, tako da lahko Dify brez težav integrirate v svojo poslovno logiko.

-## Primerjava Funkcij
-
-<table style="width: 100%;">
-  <tr>
-    <th align="center">Funkcija</th>
-    <th align="center">Dify.AI</th>
-    <th align="center">LangChain</th>
-    <th align="center">Flowise</th>
-    <th align="center">OpenAI Assistants API</th>
-  </tr>
-  <tr>
-    <td align="center">Programski pristop</td>
-    <td align="center">API + usmerjeno v aplikacije</td>
-    <td align="center">Python koda</td>
-    <td align="center">Usmerjeno v aplikacije</td>
-    <td align="center">Usmerjeno v API</td>
-  </tr>
-  <tr>
-    <td align="center">Podprti LLM-ji</td>
-    <td align="center">Bogata izbira</td>
-    <td align="center">Bogata izbira</td>
-    <td align="center">Bogata izbira</td>
-    <td align="center">Samo OpenAI</td>
-  </tr>
-  <tr>
-    <td align="center">RAG pogon</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-  </tr>
-  <tr>
-    <td align="center">Agent</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">✅</td>
-  </tr>
-  <tr>
-    <td align="center">Potek dela</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Spremljanje</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Funkcija za podjetja (SSO/nadzor dostopa)</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-    <td align="center">❌</td>
-  </tr>
-  <tr>
-    <td align="center">Lokalna namestitev</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">✅</td>
-    <td align="center">❌</td>
-  </tr>
-</table>

 ## Uporaba Dify

@@ -254,4 +187,4 @@ Zaradi zaščite vaše zasebnosti se izogibajte objavljanju varnostnih vprašanj

 ## Licenca

-To skladišče je na voljo pod [odprtokodno licenco Dify](LICENSE) , ki je v bistvu Apache 2.0 z nekaj dodatnimi omejitvami.
+To skladišče je na voljo pod [odprtokodno licenco Dify](LICENSE) , ki je v bistvu Apache 2.0 z nekaj dodatnimi omejitvami.
@@ -430,4 +430,7 @@ CREATE_TIDB_SERVICE_JOB_ENABLED=false
 # Maximum number of submitted thread count in a ThreadPool for parallel node execution
 MAX_SUBMIT_COUNT=100
 # Lockout duration in seconds
-LOGIN_LOCKOUT_DURATION=86400
+LOGIN_LOCKOUT_DURATION=86400
+
+# Prevent Clickjacking
+ALLOW_EMBED=false
@@ -37,13 +37,7 @@

 4. Create environment.

-   Dify API service uses [Poetry](https://python-poetry.org/docs/) to manage dependencies. First, you need to add the poetry shell plugin, if you don't have it already, in order to run in a virtual environment. [Note: Poetry shell is no longer a native command so you need to install the poetry plugin beforehand]
-
-   ```bash
-   poetry self add poetry-plugin-shell
-   ```
-   
-   Then, You can execute `poetry shell` to activate the environment.
+   Dify API service uses [Poetry](https://python-poetry.org/docs/) to manage dependencies. You can execute `poetry shell` to activate the environment.

 5. Install dependencies

@@ -315,8 +315,8 @@ class HttpConfig(BaseSettings):
    )

    RESPECT_XFORWARD_HEADERS_ENABLED: bool = Field(
-        description="Enable handling of X-Forwarded-For, X-Forwarded-Proto, and X-Forwarded-Port headers"
-        " when the app is behind a single trusted reverse proxy.",
+        description="Enable or disable the X-Forwarded-For Proxy Fix middleware from Werkzeug"
+        " to respect X-* headers to redirect clients",
        default=False,
    )

@@ -498,11 +498,6 @@ class AuthConfig(BaseSettings):
        default=86400,
    )

-    FORGOT_PASSWORD_LOCKOUT_DURATION: PositiveInt = Field(
-        description="Time (in seconds) a user must wait before retrying password reset after exceeding the rate limit.",
-        default=86400,
-    )
-

 class ModerationConfig(BaseSettings):
    """
@@ -1,5 +1,5 @@
 from typing import Any, Literal, Optional
-from urllib.parse import quote_plus
+from urllib.parse import parse_qsl, quote_plus

 from pydantic import Field, NonNegativeInt, PositiveFloat, PositiveInt, computed_field
 from pydantic_settings import BaseSettings
@@ -166,14 +166,28 @@ class DatabaseConfig(BaseSettings):
        default=False,
    )

-    @computed_field
+    @computed_field  # type: ignore[misc]
+    @property
    def SQLALCHEMY_ENGINE_OPTIONS(self) -> dict[str, Any]:
+        # Parse DB_EXTRAS for 'options'
+        db_extras_dict = dict(parse_qsl(self.DB_EXTRAS))
+        options = db_extras_dict.get("options", "")
+        # Always include timezone
+        timezone_opt = "-c timezone=UTC"
+        if options:
+            # Merge user options and timezone
+            merged_options = f"{options} {timezone_opt}"
+        else:
+            merged_options = timezone_opt
+
+        connect_args = {"options": merged_options}
+
        return {
            "pool_size": self.SQLALCHEMY_POOL_SIZE,
            "max_overflow": self.SQLALCHEMY_MAX_OVERFLOW,
            "pool_recycle": self.SQLALCHEMY_POOL_RECYCLE,
            "pool_pre_ping": self.SQLALCHEMY_POOL_PRE_PING,
-            "connect_args": {"options": "-c timezone=UTC"},
+            "connect_args": connect_args,
        }


@@ -9,7 +9,7 @@ class PackagingInfo(BaseSettings):

    CURRENT_VERSION: str = Field(
        description="Dify version",
-        default="0.15.3",
+        default="0.15.8",
    )

    COMMIT_SHA: str = Field(
@@ -15,7 +15,7 @@ AUDIO_EXTENSIONS.extend([ext.upper() for ext in AUDIO_EXTENSIONS])

 if dify_config.ETL_TYPE == "Unstructured":
    DOCUMENT_EXTENSIONS = ["txt", "markdown", "md", "mdx", "pdf", "html", "htm", "xlsx", "xls"]
-    DOCUMENT_EXTENSIONS.extend(("doc", "docx", "csv", "eml", "msg", "pptx", "xml", "epub"))
+    DOCUMENT_EXTENSIONS.extend(("docx", "csv", "eml", "msg", "pptx", "xml", "epub"))
    if dify_config.UNSTRUCTURED_API_URL:
        DOCUMENT_EXTENSIONS.append("ppt")
    DOCUMENT_EXTENSIONS.extend([ext.upper() for ext in DOCUMENT_EXTENSIONS])
@@ -59,9 +59,3 @@ class EmailCodeAccountDeletionRateLimitExceededError(BaseHTTPException):
    error_code = "email_code_account_deletion_rate_limit_exceeded"
    description = "Too many account deletion emails have been sent. Please try again in 5 minutes."
    code = 429
-
-
-class EmailPasswordResetLimitError(BaseHTTPException):
-    error_code = "email_password_reset_limit"
-    description = "Too many failed password reset attempts. Please try again in 24 hours."
-    code = 429
@@ -6,15 +6,9 @@ from flask_restful import Resource, reqparse  # type: ignore

 from constants.languages import languages
 from controllers.console import api
-from controllers.console.auth.error import (
-    EmailCodeError,
-    EmailPasswordResetLimitError,
-    InvalidEmailError,
-    InvalidTokenError,
-    PasswordMismatchError,
-)
+from controllers.console.auth.error import EmailCodeError, InvalidEmailError, InvalidTokenError, PasswordMismatchError
 from controllers.console.error import AccountInFreezeError, AccountNotFound, EmailSendIpLimitError
-from controllers.console.wraps import setup_required
+from controllers.console.wraps import email_password_login_enabled, setup_required
 from events.tenant_event import tenant_was_created
 from extensions.ext_database import db
 from libs.helper import email, extract_remote_ip
@@ -28,6 +22,7 @@ from services.feature_service import FeatureService

 class ForgotPasswordSendEmailApi(Resource):
    @setup_required
+    @email_password_login_enabled
    def post(self):
        parser = reqparse.RequestParser()
        parser.add_argument("email", type=email, required=True, location="json")
@@ -59,6 +54,7 @@ class ForgotPasswordSendEmailApi(Resource):

 class ForgotPasswordCheckApi(Resource):
    @setup_required
+    @email_password_login_enabled
    def post(self):
        parser = reqparse.RequestParser()
        parser.add_argument("email", type=str, required=True, location="json")
@@ -68,10 +64,6 @@ class ForgotPasswordCheckApi(Resource):

        user_email = args["email"]

-        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(args["email"])
-        if is_forgot_password_error_rate_limit:
-            raise EmailPasswordResetLimitError()
-
        token_data = AccountService.get_reset_password_data(args["token"])
        if token_data is None:
            raise InvalidTokenError()
@@ -80,15 +72,22 @@ class ForgotPasswordCheckApi(Resource):
            raise InvalidEmailError()

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

-        AccountService.reset_forgot_password_error_rate_limit(args["email"])
-        return {"is_valid": True, "email": token_data.get("email")}
+        # Verified, revoke the first token
+        AccountService.revoke_reset_password_token(args["token"])
+
+        # Refresh token data by generating a new token
+        _, new_token = AccountService.generate_reset_password_token(
+            user_email, code=args["code"], additional_data={"phase": "reset"}
+        )
+
+        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}


 class ForgotPasswordResetApi(Resource):
    @setup_required
+    @email_password_login_enabled
    def post(self):
        parser = reqparse.RequestParser()
        parser.add_argument("token", type=str, required=True, nullable=False, location="json")
@@ -107,6 +106,9 @@ class ForgotPasswordResetApi(Resource):

        if reset_data is None:
            raise InvalidTokenError()
+        # Must use token in reset phase
+        if reset_data.get("phase", "") != "reset":
+            raise InvalidTokenError()

        AccountService.revoke_reset_password_token(token)

@@ -22,7 +22,7 @@ from controllers.console.error import (
    EmailSendIpLimitError,
    NotAllowedCreateWorkspace,
 )
-from controllers.console.wraps import setup_required
+from controllers.console.wraps import email_password_login_enabled, setup_required
 from events.tenant_event import tenant_was_created
 from libs.helper import email, extract_remote_ip
 from libs.password import valid_password
@@ -38,6 +38,7 @@ class LoginApi(Resource):
    """Resource for user login."""

    @setup_required
+    @email_password_login_enabled
    def post(self):
        """Authenticate user and login."""
        parser = reqparse.RequestParser()
@@ -110,6 +111,7 @@ class LogoutApi(Resource):

 class ResetPasswordSendEmailApi(Resource):
    @setup_required
+    @email_password_login_enabled
    def post(self):
        parser = reqparse.RequestParser()
        parser.add_argument("email", type=email, required=True, location="json")
@@ -154,3 +154,16 @@ def enterprise_license_required(view):
        return view(*args, **kwargs)

    return decorated
+
+
+def email_password_login_enabled(view):
+    @wraps(view)
+    def decorated(*args, **kwargs):
+        features = FeatureService.get_system_features()
+        if features.enable_email_password_login:
+            return view(*args, **kwargs)
+
+        # otherwise, return 403
+        abort(403)
+
+    return decorated
@@ -50,8 +50,8 @@ class EnterpriseWorkspaceNoOwnerEmail(Resource):
            "plan": tenant.plan,
            "status": tenant.status,
            "custom_config": json.loads(tenant.custom_config) if tenant.custom_config else {},
-            "created_at": tenant.created_at.isoformat() + "Z" if tenant.created_at else None,
-            "updated_at": tenant.updated_at.isoformat() + "Z" if tenant.updated_at else None,
+            "created_at": tenant.created_at.isoformat() if tenant.created_at else None,
+            "updated_at": tenant.updated_at.isoformat() if tenant.updated_at else None,
        }

        return {
@@ -104,7 +104,6 @@ class CotAgentRunner(BaseAgentRunner, ABC):

            # 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,
@@ -84,7 +84,6 @@ class FunctionCallAgentRunner(BaseAgentRunner):

            # 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,
@@ -140,7 +140,9 @@ class AdvancedChatAppGenerator(MessageBasedAppGenerator):
            app_config=app_config,
            file_upload_config=file_extra_config,
            conversation_id=conversation.id if conversation else None,
-            inputs=self._prepare_user_inputs(
+            inputs=conversation.inputs
+            if conversation
+            else self._prepare_user_inputs(
                user_inputs=inputs, variables=app_config.variables, tenant_id=app_model.tenant_id
            ),
            query=query,
@@ -148,7 +148,9 @@ class AgentChatAppGenerator(MessageBasedAppGenerator):
            model_conf=ModelConfigConverter.convert(app_config),
            file_upload_config=file_extra_config,
            conversation_id=conversation.id if conversation else None,
-            inputs=self._prepare_user_inputs(
+            inputs=conversation.inputs
+            if conversation
+            else self._prepare_user_inputs(
                user_inputs=inputs, variables=app_config.variables, tenant_id=app_model.tenant_id
            ),
            query=query,
@@ -55,20 +55,6 @@ class AgentChatAppRunner(AppRunner):
        query = application_generate_entity.query
        files = application_generate_entity.files

-        # Pre-calculate the number of tokens of the prompt messages,
-        # and return the rest number of tokens by model context token size limit and max token size limit.
-        # If the rest number of tokens is not enough, raise exception.
-        # Include: prompt template, inputs, query(optional), files(optional)
-        # Not Include: memory, external data, dataset context
-        self.get_pre_calculate_rest_tokens(
-            app_record=app_record,
-            model_config=application_generate_entity.model_conf,
-            prompt_template_entity=app_config.prompt_template,
-            inputs=inputs,
-            files=files,
-            query=query,
-        )
-
        memory = None
        if application_generate_entity.conversation_id:
            # get memory of conversation (read-only)
@@ -15,10 +15,8 @@ from core.app.features.annotation_reply.annotation_reply import AnnotationReplyF
 from core.app.features.hosting_moderation.hosting_moderation import HostingModerationFeature
 from core.external_data_tool.external_data_fetch import ExternalDataFetch
 from core.memory.token_buffer_memory import TokenBufferMemory
-from core.model_manager import ModelInstance
 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
 from core.model_runtime.entities.message_entities import AssistantPromptMessage, PromptMessage
-from core.model_runtime.entities.model_entities import ModelPropertyKey
 from core.model_runtime.errors.invoke import InvokeBadRequestError
 from core.moderation.input_moderation import InputModeration
 from core.prompt.advanced_prompt_transform import AdvancedPromptTransform
@@ -31,106 +29,6 @@ if TYPE_CHECKING:


 class AppRunner:
-    def get_pre_calculate_rest_tokens(
-        self,
-        app_record: App,
-        model_config: ModelConfigWithCredentialsEntity,
-        prompt_template_entity: PromptTemplateEntity,
-        inputs: Mapping[str, str],
-        files: Sequence["File"],
-        query: Optional[str] = None,
-    ) -> int:
-        """
-        Get pre calculate rest tokens
-        :param app_record: app record
-        :param model_config: model config entity
-        :param prompt_template_entity: prompt template entity
-        :param inputs: inputs
-        :param files: files
-        :param query: query
-        :return:
-        """
-        # Invoke model
-        model_instance = ModelInstance(
-            provider_model_bundle=model_config.provider_model_bundle, model=model_config.model
-        )
-
-        model_context_tokens = model_config.model_schema.model_properties.get(ModelPropertyKey.CONTEXT_SIZE)
-
-        max_tokens = 0
-        for parameter_rule in model_config.model_schema.parameter_rules:
-            if parameter_rule.name == "max_tokens" or (
-                parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
-            ):
-                max_tokens = (
-                    model_config.parameters.get(parameter_rule.name)
-                    or model_config.parameters.get(parameter_rule.use_template or "")
-                ) or 0
-
-        if model_context_tokens is None:
-            return -1
-
-        if max_tokens is None:
-            max_tokens = 0
-
-        # get prompt messages without memory and context
-        prompt_messages, stop = self.organize_prompt_messages(
-            app_record=app_record,
-            model_config=model_config,
-            prompt_template_entity=prompt_template_entity,
-            inputs=inputs,
-            files=files,
-            query=query,
-        )
-
-        prompt_tokens = model_instance.get_llm_num_tokens(prompt_messages)
-
-        rest_tokens: int = model_context_tokens - max_tokens - prompt_tokens
-        if rest_tokens < 0:
-            raise InvokeBadRequestError(
-                "Query or prefix prompt is too long, you can reduce the prefix prompt, "
-                "or shrink the max token, or switch to a llm with a larger token limit size."
-            )
-
-        return rest_tokens
-
-    def recalc_llm_max_tokens(
-        self, model_config: ModelConfigWithCredentialsEntity, prompt_messages: list[PromptMessage]
-    ):
-        # recalc max_tokens if sum(prompt_token +  max_tokens) over model token limit
-        model_instance = ModelInstance(
-            provider_model_bundle=model_config.provider_model_bundle, model=model_config.model
-        )
-
-        model_context_tokens = model_config.model_schema.model_properties.get(ModelPropertyKey.CONTEXT_SIZE)
-
-        max_tokens = 0
-        for parameter_rule in model_config.model_schema.parameter_rules:
-            if parameter_rule.name == "max_tokens" or (
-                parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
-            ):
-                max_tokens = (
-                    model_config.parameters.get(parameter_rule.name)
-                    or model_config.parameters.get(parameter_rule.use_template or "")
-                ) or 0
-
-        if model_context_tokens is None:
-            return -1
-
-        if max_tokens is None:
-            max_tokens = 0
-
-        prompt_tokens = model_instance.get_llm_num_tokens(prompt_messages)
-
-        if prompt_tokens + max_tokens > model_context_tokens:
-            max_tokens = max(model_context_tokens - prompt_tokens, 16)
-
-            for parameter_rule in model_config.model_schema.parameter_rules:
-                if parameter_rule.name == "max_tokens" or (
-                    parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
-                ):
-                    model_config.parameters[parameter_rule.name] = max_tokens
-
    def organize_prompt_messages(
        self,
        app_record: App,
@@ -141,7 +141,9 @@ class ChatAppGenerator(MessageBasedAppGenerator):
            model_conf=ModelConfigConverter.convert(app_config),
            file_upload_config=file_extra_config,
            conversation_id=conversation.id if conversation else None,
-            inputs=self._prepare_user_inputs(
+            inputs=conversation.inputs
+            if conversation
+            else self._prepare_user_inputs(
                user_inputs=inputs, variables=app_config.variables, tenant_id=app_model.tenant_id
            ),
            query=query,
@@ -50,20 +50,6 @@ class ChatAppRunner(AppRunner):
        query = application_generate_entity.query
        files = application_generate_entity.files

-        # Pre-calculate the number of tokens of the prompt messages,
-        # and return the rest number of tokens by model context token size limit and max token size limit.
-        # If the rest number of tokens is not enough, raise exception.
-        # Include: prompt template, inputs, query(optional), files(optional)
-        # Not Include: memory, external data, dataset context
-        self.get_pre_calculate_rest_tokens(
-            app_record=app_record,
-            model_config=application_generate_entity.model_conf,
-            prompt_template_entity=app_config.prompt_template,
-            inputs=inputs,
-            files=files,
-            query=query,
-        )
-
        memory = None
        if application_generate_entity.conversation_id:
            # get memory of conversation (read-only)
@@ -194,9 +180,6 @@ class ChatAppRunner(AppRunner):
        if hosting_moderation_result:
            return

-        # Re-calculate the max tokens if sum(prompt_token +  max_tokens) over model token limit
-        self.recalc_llm_max_tokens(model_config=application_generate_entity.model_conf, prompt_messages=prompt_messages)
-
        # Invoke model
        model_instance = ModelInstance(
            provider_model_bundle=application_generate_entity.model_conf.provider_model_bundle,
@@ -43,20 +43,6 @@ class CompletionAppRunner(AppRunner):
        query = application_generate_entity.query
        files = application_generate_entity.files

-        # Pre-calculate the number of tokens of the prompt messages,
-        # and return the rest number of tokens by model context token size limit and max token size limit.
-        # If the rest number of tokens is not enough, raise exception.
-        # Include: prompt template, inputs, query(optional), files(optional)
-        # Not Include: memory, external data, dataset context
-        self.get_pre_calculate_rest_tokens(
-            app_record=app_record,
-            model_config=application_generate_entity.model_conf,
-            prompt_template_entity=app_config.prompt_template,
-            inputs=inputs,
-            files=files,
-            query=query,
-        )
-
        # organize all inputs and template to prompt messages
        # Include: prompt template, inputs, query(optional), files(optional)
        prompt_messages, stop = self.organize_prompt_messages(
@@ -152,9 +138,6 @@ class CompletionAppRunner(AppRunner):
        if hosting_moderation_result:
            return

-        # Re-calculate the max tokens if sum(prompt_token +  max_tokens) over model token limit
-        self.recalc_llm_max_tokens(model_config=application_generate_entity.model_conf, prompt_messages=prompt_messages)
-
        # Invoke model
        model_instance = ModelInstance(
            provider_model_bundle=application_generate_entity.model_conf.provider_model_bundle,
@@ -842,4 +842,4 @@ class WorkflowCycleManage:
        if node_execution_id not in self._workflow_node_executions:
            raise ValueError(f"Workflow node execution not found: {node_execution_id}")
        cached_workflow_node_execution = self._workflow_node_executions[node_execution_id]
-        return session.merge(cached_workflow_node_execution)
+        return cached_workflow_node_execution
@@ -26,7 +26,7 @@ class TokenBufferMemory:
        self.model_instance = model_instance

    def get_history_prompt_messages(
-        self, max_token_limit: int = 2000, message_limit: Optional[int] = None
+        self, max_token_limit: int = 100000, message_limit: Optional[int] = None
    ) -> Sequence[PromptMessage]:
        """
        Get history prompt messages.
@@ -30,6 +30,11 @@ from core.model_runtime.model_providers.__base.ai_model import AIModel

 logger = logging.getLogger(__name__)

+HTML_THINKING_TAG = (
+    '<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> '
+    "<summary> Thinking... </summary>"
+)
+

 class LargeLanguageModel(AIModel):
    """
@@ -403,7 +408,7 @@ if you are not sure about the structure.
    def _wrap_thinking_by_reasoning_content(self, delta: dict, is_reasoning: bool) -> tuple[str, bool]:
        """
        If the reasoning response is from delta.get("reasoning_content"), we wrap
-        it with HTML think tag.
+        it with HTML details tag.

        :param delta: delta dictionary from LLM streaming response
        :param is_reasoning: is reasoning
@@ -415,17 +420,25 @@ if you are not sure about the structure.

        if reasoning_content:
            if not is_reasoning:
-                content = "<think>\n" + reasoning_content
+                content = HTML_THINKING_TAG + reasoning_content
                is_reasoning = True
            else:
                content = reasoning_content
-        elif is_reasoning and content:
-            # do not end reasoning when content is empty
-            # there may be more reasoning_content later that follows previous reasoning closely
-            content = "\n</think>" + content
+        elif is_reasoning:
+            content = "</details>" + content
            is_reasoning = False
        return content, is_reasoning

+    def _wrap_thinking_by_tag(self, content: str) -> str:
+        """
+        if the reasoning response is a <think>...</think> block from delta.get("content"),
+        we replace <think> to <detail>.
+
+        :param content: delta.get("content")
+        :return: processed_content
+        """
+        return content.replace("<think>", HTML_THINKING_TAG).replace("</think>", "</details>")
+
    def _invoke_result_generator(
        self,
        model: str,
@@ -0,0 +1,115 @@
+model: us.anthropic.claude-3-7-sonnet-20250219-v1:0
+label:
+  en_US: Claude 3.7 Sonnet(US.Cross Region Inference)
+icon: icon_s_en.svg
+model_type: llm
+features:
+  - agent-thought
+  - vision
+  - tool-call
+  - stream-tool-call
+model_properties:
+  mode: chat
+  context_size: 200000
+# docs: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
+parameter_rules:
+  - name: enable_cache
+    label:
+      zh_Hans: 启用提示缓存
+      en_US: Enable Prompt Cache
+    type: boolean
+    required: false
+    default: true
+    help:
+      zh_Hans: 启用提示缓存可以提高性能并降低成本。Claude 3.7 Sonnet支持在system、messages和tools字段中使用缓存检查点。
+      en_US: Enable prompt caching to improve performance and reduce costs. Claude 3.7 Sonnet supports cache checkpoints in system, messages, and tools fields.
+  - name: reasoning_type
+    label:
+      zh_Hans: 推理配置
+      en_US: Reasoning Type
+    type: boolean
+    required: false
+    default: false
+    placeholder:
+      zh_Hans: 设置推理配置
+      en_US: Set reasoning configuration
+    help:
+      zh_Hans: 控制模型的推理能力。启用时，temperature将固定为1且top_p将被禁用。
+      en_US: Controls the model's reasoning capability. When enabled, temperature will be fixed to 1 and top_p will be disabled.
+  - name: reasoning_budget
+    show_on:
+      - variable: reasoning_type
+        value: true
+    label:
+      zh_Hans: 推理预算
+      en_US: Reasoning Budget
+    type: int
+    default: 1024
+    min: 0
+    max: 128000
+    help:
+      zh_Hans: 推理的预算限制（最小1024），必须小于max_tokens。仅在推理类型为enabled时可用。
+      en_US: Budget limit for reasoning (minimum 1024), must be less than max_tokens. Only available when reasoning type is enabled.
+
+  - name: max_tokens
+    use_template: max_tokens
+    required: true
+    label:
+      zh_Hans: 最大token数
+      en_US: Max Tokens
+    type: int
+    default: 8192
+    min: 1
+    max: 128000
+    help:
+      zh_Hans: 停止前生成的最大令牌数。请注意，Anthropic Claude 模型可能会在达到 max_tokens 的值之前停止生成令牌。不同的 Anthropic Claude 模型对此参数具有不同的最大值。
+      en_US: The maximum number of tokens to generate before stopping. Note that Anthropic Claude models might stop generating tokens before reaching the value of max_tokens. Different Anthropic Claude models have different maximum values for this parameter.
+  - name: temperature
+    use_template: temperature
+    required: false
+    label:
+      zh_Hans: 模型温度
+      en_US: Model Temperature
+    type: float
+    default: 1
+    min: 0.0
+    max: 1.0
+    help:
+      zh_Hans: 生成内容的随机性。当推理功能启用时，该值将被固定为1。
+      en_US: The amount of randomness injected into the response. When reasoning is enabled, this value will be fixed to 1.
+  - name: top_p
+    show_on:
+      - variable: reasoning_type
+        value: disabled
+    use_template: top_p
+    label:
+      zh_Hans: Top P
+      en_US: Top P
+    required: false
+    type: float
+    default: 0.999
+    min: 0.000
+    max: 1.000
+    help:
+      zh_Hans: 在核采样中的概率阈值。当推理功能启用时，该参数将被禁用。
+      en_US: The probability threshold in nucleus sampling. When reasoning is enabled, this parameter will be disabled.
+  - name: top_k
+    label:
+      zh_Hans: 取样数量
+      en_US: Top k
+    required: false
+    type: int
+    default: 0
+    min: 0
+    # tip docs from aws has error, max value is 500
+    max: 500
+    help:
+      zh_Hans: 对于每个后续标记，仅从前 K 个选项中进行采样。使用 top_k 删除长尾低概率响应。
+      en_US: Only sample from the top K options for each subsequent token. Use top_k to remove long tail low probability responses.
+  - name: response_format
+    use_template: response_format
+pricing:
+  input: '0.003'
+  output: '0.015'
+  unit: '0.001'
+  currency: USD
@@ -58,6 +58,7 @@ class BedrockLargeLanguageModel(LargeLanguageModel):
    # TODO There is invoke issue: context limit on Cohere Model, will add them after fixed.
    CONVERSE_API_ENABLED_MODEL_INFO = [
        {"prefix": "anthropic.claude-v2", "support_system_prompts": True, "support_tool_use": False},
+        {"prefix": "us.deepseek", "support_system_prompts": True, "support_tool_use": False},
        {"prefix": "anthropic.claude-v1", "support_system_prompts": True, "support_tool_use": False},
        {"prefix": "us.anthropic.claude-3", "support_system_prompts": True, "support_tool_use": True},
        {"prefix": "eu.anthropic.claude-3", "support_system_prompts": True, "support_tool_use": True},
@@ -0,0 +1,63 @@
+model: us.deepseek.r1-v1:0
+label:
+  en_US: DeepSeek-R1(US.Cross Region Inference)
+icon: icon_s_en.svg
+model_type: llm
+features:
+  - agent-thought
+  - vision
+  - tool-call
+  - stream-tool-call
+model_properties:
+  mode: chat
+  context_size: 32768
+parameter_rules:
+  - name: max_tokens
+    use_template: max_tokens
+    required: true
+    label:
+      zh_Hans: 最大token数
+      en_US: Max Tokens
+    type: int
+    default: 8192
+    min: 1
+    max: 128000
+    help:
+      zh_Hans: 停止前生成的最大令牌数。
+      en_US: The maximum number of tokens to generate before stopping.
+  - name: temperature
+    use_template: temperature
+    required: false
+    label:
+      zh_Hans: 模型温度
+      en_US: Model Temperature
+    type: float
+    default: 1
+    min: 0.0
+    max: 1.0
+    help:
+      zh_Hans: 生成内容的随机性。当推理功能启用时，该值将被固定为1。
+      en_US: The amount of randomness injected into the response. When reasoning is enabled, this value will be fixed to 1.
+  - name: top_p
+    show_on:
+      - variable: reasoning_type
+        value: disabled
+    use_template: top_p
+    label:
+      zh_Hans: Top P
+      en_US: Top P
+    required: false
+    type: float
+    default: 0.999
+    min: 0.000
+    max: 1.000
+    help:
+      zh_Hans: 在核采样中的概率阈值。当推理功能启用时，该参数将被禁用。
+      en_US: The probability threshold in nucleus sampling. When reasoning is enabled, this parameter will be disabled.
+  - name: response_format
+    use_template: response_format
+pricing:
+  input: '0.001'
+  output: '0.005'
+  unit: '0.001'
+  currency: USD
@@ -19,8 +19,8 @@ class GoogleProvider(ModelProvider):
        try:
            model_instance = self.get_model_instance(ModelType.LLM)

-            # Use `gemini-pro` model for validate,
-            model_instance.validate_credentials(model="gemini-pro", credentials=credentials)
+            # Use `gemini-2.0-flash` model for validate,
+            model_instance.validate_credentials(model="gemini-2.0-flash", credentials=credentials)
        except CredentialsValidateFailedError as ex:
            raise ex
        except Exception as ex:
@@ -1,6 +1,5 @@
 - gemini-2.0-flash-001
 - gemini-2.0-flash-exp
- gemini-2.0-flash-lite-preview-02-05
 - gemini-2.0-pro-exp-02-05
 - gemini-2.0-flash-thinking-exp-1219
 - gemini-2.0-flash-thinking-exp-01-21
@@ -20,5 +19,3 @@
 - gemini-exp-1206
 - gemini-exp-1121
 - gemini-exp-1114
- gemini-pro
- gemini-pro-vision
@@ -1,41 +0,0 @@
-model: gemini-2.0-flash-lite-preview-02-05
-label:
-  en_US: Gemini 2.0 Flash Lite Preview 0205
-model_type: llm
-features:
-  - agent-thought
-  - vision
-  - tool-call
-  - stream-tool-call
-  - document
-  - video
-  - audio
-model_properties:
-  mode: chat
-  context_size: 1048576
-parameter_rules:
-  - name: temperature
-    use_template: temperature
-  - name: top_p
-    use_template: top_p
-  - name: top_k
-    label:
-      zh_Hans: 取样数量
-      en_US: Top k
-    type: int
-    help:
-      zh_Hans: 仅从每个后续标记的前 K 个选项中采样。
-      en_US: Only sample from the top K options for each subsequent token.
-    required: false
-  - name: max_output_tokens
-    use_template: max_tokens
-    default: 8192
-    min: 1
-    max: 8192
-  - name: json_schema
-    use_template: json_schema
-pricing:
-  input: '0.00'
-  output: '0.00'
-  unit: '0.000001'
-  currency: USD
@@ -1,35 +0,0 @@
-model: gemini-pro-vision
-label:
-  en_US: Gemini Pro Vision
-model_type: llm
-features:
-  - vision
-model_properties:
-  mode: chat
-  context_size: 12288
-parameter_rules:
-  - name: temperature
-    use_template: temperature
-  - name: top_p
-    use_template: top_p
-  - name: top_k
-    label:
-      zh_Hans: 取样数量
-      en_US: Top k
-    type: int
-    help:
-      zh_Hans: 仅从每个后续标记的前 K 个选项中采样。
-      en_US: Only sample from the top K options for each subsequent token.
-    required: false
-  - name: max_tokens_to_sample
-    use_template: max_tokens
-    required: true
-    default: 4096
-    min: 1
-    max: 4096
-pricing:
-  input: '0.00'
-  output: '0.00'
-  unit: '0.000001'
-  currency: USD
-deprecated: true
@@ -1,39 +0,0 @@
-model: gemini-pro
-label:
-  en_US: Gemini Pro
-model_type: llm
-features:
-  - agent-thought
-  - tool-call
-  - stream-tool-call
-model_properties:
-  mode: chat
-  context_size: 30720
-parameter_rules:
-  - name: temperature
-    use_template: temperature
-  - name: top_p
-    use_template: top_p
-  - name: top_k
-    label:
-      zh_Hans: 取样数量
-      en_US: Top k
-    type: int
-    help:
-      zh_Hans: 仅从每个后续标记的前 K 个选项中采样。
-      en_US: Only sample from the top K options for each subsequent token.
-    required: false
-  - name: max_tokens_to_sample
-    use_template: max_tokens
-    required: true
-    default: 2048
-    min: 1
-    max: 2048
-  - name: response_format
-    use_template: response_format
-pricing:
-  input: '0.00'
-  output: '0.00'
-  unit: '0.000001'
-  currency: USD
-deprecated: true
@@ -367,6 +367,7 @@ class OllamaLargeLanguageModel(LargeLanguageModel):

                # transform assistant message to prompt message
                text = chunk_json["response"]
+            text = self._wrap_thinking_by_tag(text)

            assistant_prompt_message = AssistantPromptMessage(content=text)

@@ -1,3 +1,4 @@
+- gpt-4.1
 - o1
 - o1-2024-12-17
 - o1-mini
@@ -1,38 +1,43 @@
-model: deepseek-v3
+model: gpt-4.1
 label:
-  zh_Hans: DeepSeek-V3
-  en_US: DeepSeek-V3
+  zh_Hans: gpt-4.1
+  en_US: gpt-4.1
 model_type: llm
 features:
+  - multi-tool-call
  - agent-thought
+  - stream-tool-call
+  - vision
 model_properties:
  mode: chat
-  context_size: 64000
+  context_size: 1047576
 parameter_rules:
  - name: temperature
    use_template: temperature
-  - name: max_tokens
-    use_template: max_tokens
-    type: int
-    default: 512
-    min: 1
-    max: 4096
-    help:
-      zh_Hans: 指定生成结果长度的上限。如果生成结果截断，可以调大该参数。
-      en_US: Specifies the upper limit on the length of generated results. If the generated results are truncated, you can increase this parameter.
  - name: top_p
    use_template: top_p
-  - name: top_k
-    label:
-      zh_Hans: 取样数量
-      en_US: Top k
-    type: int
-    help:
-      zh_Hans: 仅从每个后续标记的前 K 个选项中采样。
-      en_US: Only sample from the top K options for each subsequent token.
-    required: false
+  - name: presence_penalty
+    use_template: presence_penalty
  - name: frequency_penalty
    use_template: frequency_penalty
+  - name: max_tokens
+    use_template: max_tokens
+    default: 512
+    min: 1
+    max: 32768
+  - name: reasoning_effort
+    label:
+      zh_Hans: 推理工作
+      en_US: Reasoning Effort
+    type: string
+    help:
+      zh_Hans: 限制推理模型的推理工作
+      en_US: Constrains effort on reasoning for reasoning models
+    required: false
+    options:
+      - low
+      - medium
+      - high
  - name: response_format
    label:
      zh_Hans: 回复格式
@@ -45,8 +50,11 @@ parameter_rules:
    options:
      - text
      - json_object
+      - json_schema
+  - name: json_schema
+    use_template: json_schema
 pricing:
-  input: "0.002"
-  output: "0.008"
-  unit: "0.001"
-  currency: RMB
+  input: '2.00'
+  output: '8.00'
+  unit: '0.000001'
+  currency: USD
@@ -1049,6 +1049,9 @@ class OpenAILargeLanguageModel(_CommonOpenAI, LargeLanguageModel):
        """Calculate num tokens for gpt-3.5-turbo and gpt-4 with tiktoken package.

        Official documentation: https://github.com/openai/openai-cookbook/blob/main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
+        if not messages and not tools:
+            return 0
+
        if model.startswith("ft:"):
            model = model.split(":")[1]

@@ -1057,18 +1060,18 @@ class OpenAILargeLanguageModel(_CommonOpenAI, LargeLanguageModel):
            model = "gpt-4o"

        try:
-            encoding = tiktoken.encoding_for_model(model)
-        except KeyError:
-            logger.warning("Warning: model not found. Using cl100k_base encoding.")
-            model = "cl100k_base"
            encoding = tiktoken.get_encoding(model)
+        except (KeyError, ValueError) as e:
+            logger.warning("Warning: model not found. Using cl100k_base encoding.")
+            encoding_name = "cl100k_base"
+            encoding = tiktoken.get_encoding(encoding_name)

        if model.startswith("gpt-3.5-turbo-0301"):
            # every message follows <im_start>{role/name}\n{content}<im_end>\n
            tokens_per_message = 4
            # if there's a name, the role is omitted
            tokens_per_name = -1
-        elif model.startswith("gpt-3.5-turbo") or model.startswith("gpt-4") or model.startswith(("o1", "o3")):
+        elif model.startswith("gpt-3.5-turbo") or model.startswith("gpt-4") or model.startswith(("o1", "o3", "o4")):
            tokens_per_message = 3
            tokens_per_name = 1
        else:
@@ -528,6 +528,7 @@ class OAIAPICompatLargeLanguageModel(_CommonOaiApiCompat, LargeLanguageModel):
                    delta_content, is_reasoning_started = self._wrap_thinking_by_reasoning_content(
                        delta, is_reasoning_started
                    )
+                    delta_content = self._wrap_thinking_by_tag(delta_content)

                    assistant_message_tool_calls = None

@@ -807,37 +808,34 @@ class OAIAPICompatLargeLanguageModel(_CommonOaiApiCompat, LargeLanguageModel):

            # calculate num tokens for function object
            num_tokens += self._get_num_tokens_by_gpt2("name")
-            if hasattr(tool, "name"):
-                num_tokens += self._get_num_tokens_by_gpt2(tool.name)
+            num_tokens += self._get_num_tokens_by_gpt2(tool.name)
            num_tokens += self._get_num_tokens_by_gpt2("description")
-            if hasattr(tool, "description"):
-                num_tokens += self._get_num_tokens_by_gpt2(tool.description)
-            if hasattr(tool, "parameters"):
-                parameters = tool.parameters
-                num_tokens += self._get_num_tokens_by_gpt2("parameters")
-                if "title" in parameters:
-                    num_tokens += self._get_num_tokens_by_gpt2("title")
-                    num_tokens += self._get_num_tokens_by_gpt2(parameters.get("title"))
-                num_tokens += self._get_num_tokens_by_gpt2("type")
-                num_tokens += self._get_num_tokens_by_gpt2(parameters.get("type"))
-                if "properties" in parameters:
-                    num_tokens += self._get_num_tokens_by_gpt2("properties")
-                    for key, value in parameters.get("properties", {}).items():
-                        num_tokens += self._get_num_tokens_by_gpt2(key)
-                        for field_key, field_value in value.items():
+            num_tokens += self._get_num_tokens_by_gpt2(tool.description)
+            parameters = tool.parameters
+            num_tokens += self._get_num_tokens_by_gpt2("parameters")
+            if "title" in parameters:
+                num_tokens += self._get_num_tokens_by_gpt2("title")
+                num_tokens += self._get_num_tokens_by_gpt2(parameters.get("title"))
+            num_tokens += self._get_num_tokens_by_gpt2("type")
+            num_tokens += self._get_num_tokens_by_gpt2(parameters.get("type"))
+            if "properties" in parameters:
+                num_tokens += self._get_num_tokens_by_gpt2("properties")
+                for key, value in parameters.get("properties").items():
+                    num_tokens += self._get_num_tokens_by_gpt2(key)
+                    for field_key, field_value in value.items():
+                        num_tokens += self._get_num_tokens_by_gpt2(field_key)
+                        if field_key == "enum":
+                            for enum_field in field_value:
+                                num_tokens += 3
+                                num_tokens += self._get_num_tokens_by_gpt2(enum_field)
+                        else:
                            num_tokens += self._get_num_tokens_by_gpt2(field_key)
-                            if field_key == "enum":
-                                for enum_field in field_value:
-                                    num_tokens += 3
-                                    num_tokens += self._get_num_tokens_by_gpt2(enum_field)
-                            else:
-                                num_tokens += self._get_num_tokens_by_gpt2(field_key)
-                                num_tokens += self._get_num_tokens_by_gpt2(str(field_value))
-                if "required" in parameters:
-                    num_tokens += self._get_num_tokens_by_gpt2("required")
-                    for required_field in parameters["required"]:
-                        num_tokens += 3
-                        num_tokens += self._get_num_tokens_by_gpt2(required_field)
+                            num_tokens += self._get_num_tokens_by_gpt2(str(field_value))
+            if "required" in parameters:
+                num_tokens += self._get_num_tokens_by_gpt2("required")
+                for required_field in parameters["required"]:
+                    num_tokens += 3
+                    num_tokens += self._get_num_tokens_by_gpt2(required_field)

        return num_tokens

@@ -430,7 +430,7 @@ class SageMakerLargeLanguageModel(LargeLanguageModel):
                type=ParameterType.INT,
                use_template="max_tokens",
                min=1,
-                max=int(credentials.get("context_length", 2048)),
+                max=credentials.get("context_length", 2048),
                default=512,
                label=I18nObject(zh_Hans="最大生成长度", en_US="Max Tokens"),
            ),
@@ -448,7 +448,7 @@ class SageMakerLargeLanguageModel(LargeLanguageModel):
        if support_vision:
            features.append(ModelFeature.VISION)

-        context_length = int(credentials.get("context_length", 2048))
+        context_length = credentials.get("context_length", 2048)

        entity = AIModelEntity(
            model=model,
@@ -59,19 +59,6 @@ model_credential_schema:
      placeholder:
        zh_Hans: 请输出你的Sagemaker推理端点
        en_US: Enter your Sagemaker Inference endpoint
-    - variable: context_length
-      show_on:
-        - variable: __model_type
-          value: llm
-      label:
-        zh_Hans: 模型上下文长度
-        en_US: Model context size
-      type: text-input
-      default: '4096'
-      required: true
-      placeholder:
-        zh_Hans: 在此输入您的模型上下文长度
-        en_US: Enter your Model context size
    - variable: audio_s3_cache_bucket
      show_on:
        - variable: __model_type
@@ -17,13 +17,6 @@
 - deepseek-ai/DeepSeek-V2.5
 - deepseek-ai/DeepSeek-V3
 - deepseek-ai/DeepSeek-Coder-V2-Instruct
- deepseek-ai/DeepSeek-R1-Distill-Llama-8B
- deepseek-ai/DeepSeek-R1-Distill-Llama-70B
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
- deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
- deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
- deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
- deepseek-ai/Janus-Pro-7B
 - THUDM/glm-4-9b-chat
 - 01-ai/Yi-1.5-34B-Chat-16K
 - 01-ai/Yi-1.5-9B-Chat-16K
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "4.3"
-  unit: "0.000001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Llama-8B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Llama-8B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Llama-8B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "0.00"
-  unit: "0.000001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "1.26"
-  unit: "0.000001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "0.70"
-  unit: "0.000001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "1.26"
-  unit: "0.000001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
-label:
-  zh_Hans: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
-  en_US: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "0.00"
-  unit: "0.000001"
-  currency: RMB
@@ -1,22 +0,0 @@
-model: deepseek-ai/Janus-Pro-7B
-label:
-  zh_Hans: deepseek-ai/Janus-Pro-7B
-  en_US: deepseek-ai/Janus-Pro-7B
-model_type: llm
-features:
-  - agent-thought
-  - vision
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.00"
-  output: "0.00"
-  unit: "0.000001"
-  currency: RMB
@@ -1,7 +1,3 @@
- deepseek-r1
- deepseek-r1-distill-qwen-14b
- deepseek-r1-distill-qwen-32b
- deepseek-v3
 - qwen-vl-max-0809
 - qwen-vl-max-0201
 - qwen-vl-max
@@ -1,21 +0,0 @@
-model: deepseek-r1-distill-qwen-14b
-label:
-  zh_Hans: DeepSeek-R1-Distill-Qwen-14B
-  en_US: DeepSeek-R1-Distill-Qwen-14B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.001"
-  output: "0.003"
-  unit: "0.001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-r1-distill-qwen-32b
-label:
-  zh_Hans: DeepSeek-R1-Distill-Qwen-32B
-  en_US: DeepSeek-R1-Distill-Qwen-32B
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 32000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.002"
-  output: "0.006"
-  unit: "0.001"
-  currency: RMB
@@ -1,21 +0,0 @@
-model: deepseek-r1
-label:
-  zh_Hans: DeepSeek-R1
-  en_US: DeepSeek-R1
-model_type: llm
-features:
-  - agent-thought
-model_properties:
-  mode: chat
-  context_size: 64000
-parameter_rules:
-  - name: max_tokens
-    use_template: max_tokens
-    min: 1
-    max: 8192
-    default: 4096
-pricing:
-  input: "0.004"
-  output: "0.016"
-  unit: '0.001'
-  currency: RMB
@@ -197,7 +197,8 @@ class TongyiLargeLanguageModel(LargeLanguageModel):
        else:
            # nothing different between chat model and completion model in tongyi
            params["messages"] = self._convert_prompt_messages_to_tongyi_messages(prompt_messages)
-            response = Generation.call(**params, result_format="message", stream=stream, incremental_output=stream)
+            response = Generation.call(**params, result_format="message", stream=stream)
+
        if stream:
            return self._handle_generate_stream_response(model, credentials, response, prompt_messages)

@@ -257,9 +258,6 @@ class TongyiLargeLanguageModel(LargeLanguageModel):
        """
        full_text = ""
        tool_calls = []
-        is_reasoning_started = False
-        # for index, response in enumerate(responses):
-        index = 0
        for index, response in enumerate(responses):
            if response.status_code not in {200, HTTPStatus.OK}:
                raise ServiceUnavailableError(
@@ -313,11 +311,7 @@ class TongyiLargeLanguageModel(LargeLanguageModel):
                    ),
                )
            else:
-                message = response.output.choices[0].message
-
-                resp_content, is_reasoning_started = self._wrap_thinking_by_reasoning_content(
-                    message, is_reasoning_started
-                )
+                resp_content = response.output.choices[0].message.content
                if not resp_content:
                    if "tool_calls" in response.output.choices[0].message:
                        tool_calls = response.output.choices[0].message["tool_calls"]
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -68,15 +68,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -68,15 +68,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -69,15 +69,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -67,15 +67,6 @@ parameter_rules:
    help:
      zh_Hans: 用于控制模型生成时的重复度。提高repetition_penalty时可以降低模型生成的重复度。1.0表示不做惩罚。
      en_US: Used to control the repeatability when generating models. Increasing repetition_penalty can reduce the duplication of model generation. 1.0 means no punishment.
-  - name: enable_search
-    type: boolean
-    default: false
-    label:
-      zh_Hans: 联网搜索
-      en_US: Web Search
-    help:
-      zh_Hans: 模型内置了互联网搜索服务，该参数控制模型在生成文本时是否参考使用互联网搜索结果。启用互联网搜索，模型会将搜索结果作为文本生成过程中的参考信息，但模型会基于其内部逻辑“自行判断”是否使用互联网搜索结果。
-      en_US: The model has a built-in Internet search service. This parameter controls whether the model refers to Internet search results when generating text. When Internet search is enabled, the model will use the search results as reference information in the text generation process, but the model will "judge" whether to use Internet search results based on its internal logic.
  - name: response_format
    use_template: response_format
 pricing:
@@ -1,22 +0,0 @@
- claude-3-haiku@20240307
- claude-3-opus@20240229
- claude-3-sonnet@20240229
- claude-3-5-sonnet-v2@20241022
- claude-3-5-sonnet@20240620
- gemini-1.0-pro-vision-001
- gemini-1.0-pro-002
- gemini-1.5-flash-001
- gemini-1.5-flash-002
- gemini-1.5-pro-001
- gemini-1.5-pro-002
- gemini-2.0-flash-001
- gemini-2.0-flash-exp
- gemini-2.0-flash-lite-preview-02-05
- gemini-2.0-flash-thinking-exp-01-21
- gemini-2.0-flash-thinking-exp-1219
- gemini-2.0-pro-exp-02-05
- gemini-exp-1114
- gemini-exp-1121
- gemini-exp-1206
- gemini-flash-experimental
- gemini-pro-experimental
@@ -5,11 +5,6 @@ model_type: llm
 features:
  - agent-thought
  - vision
-  - tool-call
-  - stream-tool-call
-  - document
-  - video
-  - audio
 model_properties:
  mode: chat
  context_size: 1048576
@@ -20,20 +15,21 @@ parameter_rules:
    use_template: top_p
  - name: top_k
    label:
-      zh_Hans: 取样数量
      en_US: Top k
    type: int
    help:
-      zh_Hans: 仅从每个后续标记的前 K 个选项中采样。
      en_US: Only sample from the top K options for each subsequent token.
    required: false
+  - name: presence_penalty
+    use_template: presence_penalty
+  - name: frequency_penalty
+    use_template: frequency_penalty
  - name: max_output_tokens
    use_template: max_tokens
+    required: true
    default: 8192
    min: 1
    max: 8192
-  - name: json_schema
-    use_template: json_schema
 pricing:
  input: '0.00'
  output: '0.00'
@@ -1,4 +1,5 @@
 import logging
+import re
 from collections.abc import Generator
 from typing import Optional

@@ -230,17 +231,6 @@ class VolcengineMaaSLargeLanguageModel(LargeLanguageModel):
            return _handle_chat_response()
        return _handle_stream_chat_response()

-    def wrap_thinking(self, delta: dict, is_reasoning: bool) -> tuple[str, bool]:
-        content = ""
-        reasoning_content = None
-        if hasattr(delta, "content"):
-            content = delta.content
-        if hasattr(delta, "reasoning_content"):
-            reasoning_content = delta.reasoning_content
-        return self._wrap_thinking_by_reasoning_content(
-            {"content": content, "reasoning_content": reasoning_content}, is_reasoning
-        )
-
    def _generate_v3(
        self,
        model: str,
@@ -263,7 +253,22 @@ class VolcengineMaaSLargeLanguageModel(LargeLanguageModel):
                content = ""
                if chunk.choices:
                    delta = chunk.choices[0].delta
-                    content, is_reasoning_started = self.wrap_thinking(delta, is_reasoning_started)
+                    if is_reasoning_started and not hasattr(delta, "reasoning_content") and not delta.content:
+                        content = ""
+                    elif hasattr(delta, "reasoning_content"):
+                        if not is_reasoning_started:
+                            is_reasoning_started = True
+                            content = "> 💭 " + delta.reasoning_content
+                        else:
+                            content = delta.reasoning_content
+
+                        if "\n" in content:
+                            content = re.sub(r"\n(?!(>|\n))", "\n> ", content)
+                    elif is_reasoning_started:
+                        content = "\n\n" + delta.content
+                        is_reasoning_started = False
+                    else:
+                        content = delta.content

                yield LLMResultChunk(
                    model=model,
@@ -328,71 +333,54 @@ class VolcengineMaaSLargeLanguageModel(LargeLanguageModel):
        """
        model_config = get_model_config(credentials)

-        if model.startswith("DeepSeek-R1"):
-            rules = [
-                ParameterRule(
-                    name="max_tokens",
-                    type=ParameterType.INT,
-                    use_template="max_tokens",
-                    min=1,
-                    max=model_config.properties.max_tokens,
-                    default=512,
-                    label=I18nObject(zh_Hans="最大生成长度", en_US="Max Tokens"),
+        rules = [
+            ParameterRule(
+                name="temperature",
+                type=ParameterType.FLOAT,
+                use_template="temperature",
+                label=I18nObject(zh_Hans="温度", en_US="Temperature"),
+            ),
+            ParameterRule(
+                name="top_p",
+                type=ParameterType.FLOAT,
+                use_template="top_p",
+                label=I18nObject(zh_Hans="Top P", en_US="Top P"),
+            ),
+            ParameterRule(
+                name="top_k", type=ParameterType.INT, min=1, default=1, label=I18nObject(zh_Hans="Top K", en_US="Top K")
+            ),
+            ParameterRule(
+                name="presence_penalty",
+                type=ParameterType.FLOAT,
+                use_template="presence_penalty",
+                label=I18nObject(
+                    en_US="Presence Penalty",
+                    zh_Hans="存在惩罚",
                ),
-            ]
-        else:
-            rules = [
-                ParameterRule(
-                    name="temperature",
-                    type=ParameterType.FLOAT,
-                    use_template="temperature",
-                    label=I18nObject(zh_Hans="温度", en_US="Temperature"),
+                min=-2.0,
+                max=2.0,
+            ),
+            ParameterRule(
+                name="frequency_penalty",
+                type=ParameterType.FLOAT,
+                use_template="frequency_penalty",
+                label=I18nObject(
+                    en_US="Frequency Penalty",
+                    zh_Hans="频率惩罚",
                ),
-                ParameterRule(
-                    name="top_p",
-                    type=ParameterType.FLOAT,
-                    use_template="top_p",
-                    label=I18nObject(zh_Hans="Top P", en_US="Top P"),
-                ),
-                ParameterRule(
-                    name="top_k",
-                    type=ParameterType.INT,
-                    min=1,
-                    default=1,
-                    label=I18nObject(zh_Hans="Top K", en_US="Top K"),
-                ),
-                ParameterRule(
-                    name="presence_penalty",
-                    type=ParameterType.FLOAT,
-                    use_template="presence_penalty",
-                    label=I18nObject(
-                        en_US="Presence Penalty",
-                        zh_Hans="存在惩罚",
-                    ),
-                    min=-2.0,
-                    max=2.0,
-                ),
-                ParameterRule(
-                    name="frequency_penalty",
-                    type=ParameterType.FLOAT,
-                    use_template="frequency_penalty",
-                    label=I18nObject(
-                        en_US="Frequency Penalty",
-                        zh_Hans="频率惩罚",
-                    ),
-                    min=-2.0,
-                    max=2.0,
-                ),
-                ParameterRule(
-                    name="max_tokens",
-                    type=ParameterType.INT,
-                    use_template="max_tokens",
-                    min=1,
-                    max=model_config.properties.max_tokens,
-                    default=512,
-                    label=I18nObject(zh_Hans="最大生成长度", en_US="Max Tokens"),
-                ),
-            ]
+                min=-2.0,
+                max=2.0,
+            ),
+            ParameterRule(
+                name="max_tokens",
+                type=ParameterType.INT,
+                use_template="max_tokens",
+                min=1,
+                max=model_config.properties.max_tokens,
+                default=512,
+                label=I18nObject(zh_Hans="最大生成长度", en_US="Max Tokens"),
+            ),
+        ]

        model_properties = {}
        model_properties[ModelPropertyKey.CONTEXT_SIZE] = model_config.properties.context_size
@@ -654,6 +654,7 @@ class XinferenceAILargeLanguageModel(LargeLanguageModel):
            if function_call:
                assistant_message_tool_calls += [self._extract_response_function_call(function_call)]

+            delta_content = self._wrap_thinking_by_tag(delta_content)
            # transform assistant message to prompt message
            assistant_prompt_message = AssistantPromptMessage(
                content=delta_content or "", tool_calls=assistant_message_tool_calls
@@ -452,9 +452,11 @@ class ProviderManager:

        provider_name_to_provider_load_balancing_model_configs_dict = defaultdict(list)
        for provider_load_balancing_config in provider_load_balancing_configs:
-            provider_name_to_provider_load_balancing_model_configs_dict[
-                provider_load_balancing_config.provider_name
-            ].append(provider_load_balancing_config)
+            (
+                provider_name_to_provider_load_balancing_model_configs_dict[
+                    provider_load_balancing_config.provider_name
+                ].append(provider_load_balancing_config)
+            )

        return provider_name_to_provider_load_balancing_model_configs_dict

@@ -1,4 +1,3 @@
-import json
 import threading
 from typing import Optional

@@ -172,7 +171,7 @@ class RetrievalService:
                vector = Vector(dataset=dataset)

                documents = vector.search_by_vector(
-                    query,
+                    cls.escape_query_for_search(query),
                    search_type="similarity_score_threshold",
                    top_k=top_k,
                    score_threshold=score_threshold,
@@ -251,7 +250,7 @@ class RetrievalService:

    @staticmethod
    def escape_query_for_search(query: str) -> str:
-        return json.dumps(query).strip('"')
+        return query.replace('"', '\\"')

    @staticmethod
    def format_retrieval_documents(documents: list[Document]) -> list[RetrievalSegments]:
@@ -9,7 +9,6 @@ from sqlalchemy import text as sql_text
 from sqlalchemy.orm import Session, declarative_base

 from configs import dify_config
-from core.rag.datasource.vdb.field import Field
 from core.rag.datasource.vdb.vector_base import BaseVector
 from core.rag.datasource.vdb.vector_factory import AbstractVectorFactory
 from core.rag.datasource.vdb.vector_type import VectorType
@@ -55,13 +54,14 @@ class TiDBVector(BaseVector):
        return Table(
            self._collection_name,
            self._orm_base.metadata,
-            Column(Field.PRIMARY_KEY.value, String(36), primary_key=True, nullable=False),
+            Column("id", String(36), primary_key=True, nullable=False),
            Column(
-                Field.VECTOR.value,
+                "vector",
                VectorType(dim),
                nullable=False,
+                comment="" if self._distance_func is None else f"hnsw(distance={self._distance_func})",
            ),
-            Column(Field.TEXT_KEY.value, TEXT, nullable=False),
+            Column("text", TEXT, nullable=False),
            Column("meta", JSON, nullable=False),
            Column("create_time", DateTime, server_default=sqlalchemy.text("CURRENT_TIMESTAMP")),
            Column(
@@ -96,7 +96,6 @@ class TiDBVector(BaseVector):
            collection_exist_cache_key = "vector_indexing_{}".format(self._collection_name)
            if redis_client.get(collection_exist_cache_key):
                return
-            tidb_dist_func = self._get_distance_func()
            with Session(self._engine) as session:
                session.begin()
                create_statement = sql_text(f"""
@@ -105,14 +104,14 @@ class TiDBVector(BaseVector):
                        text TEXT NOT NULL,
                        meta JSON NOT NULL,
                        doc_id VARCHAR(64) AS (JSON_UNQUOTE(JSON_EXTRACT(meta, '$.doc_id'))) STORED,
-                        vector VECTOR<FLOAT>({dimension}) NOT NULL,
-                        create_time DATETIME DEFAULT CURRENT_TIMESTAMP,
-                        update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
                        KEY (doc_id),
-                        VECTOR INDEX idx_vector (({tidb_dist_func}(vector))) USING HNSW
+                        vector VECTOR<FLOAT>({dimension}) NOT NULL COMMENT "hnsw(distance={self._distance_func})",
+                        create_time DATETIME DEFAULT CURRENT_TIMESTAMP,
+                        update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP
                    );
                """)
                session.execute(create_statement)
+                # tidb vector not support 'CREATE/ADD INDEX' now
                session.commit()
            redis_client.set(collection_exist_cache_key, 1, ex=3600)

@@ -195,30 +194,23 @@ class TiDBVector(BaseVector):
        )

        docs = []
-        tidb_dist_func = self._get_distance_func()
+        if self._distance_func == "l2":
+            tidb_func = "Vec_l2_distance"
+        elif self._distance_func == "cosine":
+            tidb_func = "Vec_Cosine_distance"
+        else:
+            tidb_func = "Vec_Cosine_distance"

        with Session(self._engine) as session:
-            select_statement = sql_text(f"""
-                SELECT meta, text, distance 
-                FROM (
-                  SELECT 
-                    meta,
-                    text,
-                    {tidb_dist_func}(vector, :query_vector_str) AS distance
-                  FROM {self._collection_name}
-                  ORDER BY distance ASC
-                  LIMIT :top_k
-                ) t
-                WHERE distance <= :distance
-                """)
-            res = session.execute(
-                select_statement,
-                params={
-                    "query_vector_str": query_vector_str,
-                    "distance": distance,
-                    "top_k": top_k,
-                },
+            select_statement = sql_text(
+                f"""SELECT meta, text, distance FROM (
+                        SELECT meta, text, {tidb_func}(vector, "{query_vector_str}")  as distance
+                        FROM {self._collection_name}
+                        ORDER BY distance
+                        LIMIT {top_k}
+                    ) t WHERE distance < {distance};"""
            )
+            res = session.execute(select_statement)
            results = [(row[0], row[1], row[2]) for row in res]
            for meta, text, distance in results:
                metadata = json.loads(meta)
@@ -235,16 +227,6 @@ class TiDBVector(BaseVector):
            session.execute(sql_text(f"""DROP TABLE IF EXISTS {self._collection_name};"""))
            session.commit()

-    def _get_distance_func(self) -> str:
-        match self._distance_func:
-            case "l2":
-                tidb_dist_func = "VEC_L2_DISTANCE"
-            case "cosine":
-                tidb_dist_func = "VEC_COSINE_DISTANCE"
-            case _:
-                tidb_dist_func = "VEC_COSINE_DISTANCE"
-        return tidb_dist_func
-

 class TiDBVectorFactory(AbstractVectorFactory):
    def init_vector(self, dataset: Dataset, attributes: list, embeddings: Embeddings) -> TiDBVector:
@@ -77,5 +77,4 @@
 - onebot
 - regex
 - trello
- vanna
 - fal
@@ -1,134 +0,0 @@
-from typing import Any, Union
-
-from vanna.remote import VannaDefault  # type: ignore
-
-from core.tools.entities.tool_entities import ToolInvokeMessage
-from core.tools.errors import ToolProviderCredentialValidationError
-from core.tools.tool.builtin_tool import BuiltinTool
-
-
-class VannaTool(BuiltinTool):
-    def _invoke(
-        self, user_id: str, tool_parameters: dict[str, Any]
-    ) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]:
-        """
-        invoke tools
-        """
-        # Ensure runtime and credentials
-        if not self.runtime or not self.runtime.credentials:
-            raise ToolProviderCredentialValidationError("Tool runtime or credentials are missing")
-        api_key = self.runtime.credentials.get("api_key", None)
-        if not api_key:
-            raise ToolProviderCredentialValidationError("Please input api key")
-
-        model = tool_parameters.get("model", "")
-        if not model:
-            return self.create_text_message("Please input RAG model")
-
-        prompt = tool_parameters.get("prompt", "")
-        if not prompt:
-            return self.create_text_message("Please input prompt")
-
-        url = tool_parameters.get("url", "")
-        if not url:
-            return self.create_text_message("Please input URL/Host/DSN")
-
-        db_name = tool_parameters.get("db_name", "")
-        username = tool_parameters.get("username", "")
-        password = tool_parameters.get("password", "")
-        port = tool_parameters.get("port", 0)
-
-        base_url = self.runtime.credentials.get("base_url", None)
-        vn = VannaDefault(model=model, api_key=api_key, config={"endpoint": base_url})
-
-        db_type = tool_parameters.get("db_type", "")
-        if db_type in {"Postgres", "MySQL", "Hive", "ClickHouse"}:
-            if not db_name:
-                return self.create_text_message("Please input database name")
-            if not username:
-                return self.create_text_message("Please input username")
-            if port < 1:
-                return self.create_text_message("Please input port")
-
-        schema_sql = "SELECT * FROM INFORMATION_SCHEMA.COLUMNS"
-        match db_type:
-            case "SQLite":
-                schema_sql = "SELECT type, sql FROM sqlite_master WHERE sql is not null"
-                vn.connect_to_sqlite(url)
-            case "Postgres":
-                vn.connect_to_postgres(host=url, dbname=db_name, user=username, password=password, port=port)
-            case "DuckDB":
-                vn.connect_to_duckdb(url=url)
-            case "SQLServer":
-                vn.connect_to_mssql(url)
-            case "MySQL":
-                vn.connect_to_mysql(host=url, dbname=db_name, user=username, password=password, port=port)
-            case "Oracle":
-                vn.connect_to_oracle(user=username, password=password, dsn=url)
-            case "Hive":
-                vn.connect_to_hive(host=url, dbname=db_name, user=username, password=password, port=port)
-            case "ClickHouse":
-                vn.connect_to_clickhouse(host=url, dbname=db_name, user=username, password=password, port=port)
-
-        enable_training = tool_parameters.get("enable_training", False)
-        reset_training_data = tool_parameters.get("reset_training_data", False)
-        if enable_training:
-            if reset_training_data:
-                existing_training_data = vn.get_training_data()
-                if len(existing_training_data) > 0:
-                    for _, training_data in existing_training_data.iterrows():
-                        vn.remove_training_data(training_data["id"])
-
-            ddl = tool_parameters.get("ddl", "")
-            question = tool_parameters.get("question", "")
-            sql = tool_parameters.get("sql", "")
-            memos = tool_parameters.get("memos", "")
-            training_metadata = tool_parameters.get("training_metadata", False)
-
-            if training_metadata:
-                if db_type == "SQLite":
-                    df_ddl = vn.run_sql(schema_sql)
-                    for ddl in df_ddl["sql"].to_list():
-                        vn.train(ddl=ddl)
-                else:
-                    df_information_schema = vn.run_sql(schema_sql)
-                    plan = vn.get_training_plan_generic(df_information_schema)
-                    vn.train(plan=plan)
-
-            if ddl:
-                vn.train(ddl=ddl)
-
-            if sql:
-                if question:
-                    vn.train(question=question, sql=sql)
-                else:
-                    vn.train(sql=sql)
-            if memos:
-                vn.train(documentation=memos)
-
-        #########################################################################################
-        # Due to CVE-2024-5565, we have to disable the chart generation feature
-        # The Vanna library uses a prompt function to present the user with visualized results,
-        # it is possible to alter the prompt using prompt injection and run arbitrary Python code
-        # instead of the intended visualization code.
-        # Specifically - allowing external input to the library’s “ask” method
-        # with "visualize" set to True (default behavior) leads to remote code execution.
-        # Affected versions: <= 0.5.5
-        #########################################################################################
-        allow_llm_to_see_data = tool_parameters.get("allow_llm_to_see_data", False)
-        res = vn.ask(
-            prompt, print_results=False, auto_train=True, visualize=False, allow_llm_to_see_data=allow_llm_to_see_data
-        )
-
-        result = []
-
-        if res is not None:
-            result.append(self.create_text_message(res[0]))
-            if len(res) > 1 and res[1] is not None:
-                result.append(self.create_text_message(res[1].to_markdown()))
-            if len(res) > 2 and res[2] is not None:
-                result.append(
-                    self.create_blob_message(blob=res[2].to_image(format="svg"), meta={"mime_type": "image/svg+xml"})
-                )
-
-        return result
@@ -1,213 +0,0 @@
-identity:
-  name: vanna
-  author: QCTC
-  label:
-    en_US: Vanna.AI
-    zh_Hans: Vanna.AI
-description:
-  human:
-    en_US: The fastest way to get actionable insights from your database just by asking questions.
-    zh_Hans: 一个基于大模型和RAG的Text2SQL工具。
-  llm: A tool for converting text to SQL.
-parameters:
-  - name: prompt
-    type: string
-    required: true
-    label:
-      en_US: Prompt
-      zh_Hans: 提示词
-      pt_BR: Prompt
-    human_description:
-      en_US: used for generating SQL
-      zh_Hans: 用于生成SQL
-    llm_description: key words for generating SQL
-    form: llm
-  - name: model
-    type: string
-    required: true
-    label:
-      en_US: RAG Model
-      zh_Hans: RAG模型
-    human_description:
-      en_US: RAG Model for your database DDL
-      zh_Hans: 存储数据库训练数据的RAG模型
-    llm_description: RAG Model for generating SQL
-    form: llm
-  - name: db_type
-    type: select
-    required: true
-    options:
-      - value: SQLite
-        label:
-          en_US: SQLite
-          zh_Hans: SQLite
-      - value: Postgres
-        label:
-          en_US: Postgres
-          zh_Hans: Postgres
-      - value: DuckDB
-        label:
-          en_US: DuckDB
-          zh_Hans: DuckDB
-      - value: SQLServer
-        label:
-          en_US: Microsoft SQL Server
-          zh_Hans: 微软 SQL Server
-      - value: MySQL
-        label:
-          en_US: MySQL
-          zh_Hans: MySQL
-      - value: Oracle
-        label:
-          en_US: Oracle
-          zh_Hans: Oracle
-      - value: Hive
-        label:
-          en_US: Hive
-          zh_Hans: Hive
-      - value: ClickHouse
-        label:
-          en_US: ClickHouse
-          zh_Hans: ClickHouse
-    default: SQLite
-    label:
-      en_US: DB Type
-      zh_Hans: 数据库类型
-    human_description:
-      en_US: Database type.
-      zh_Hans: 选择要链接的数据库类型。
-    form: form
-  - name: url
-    type: string
-    required: true
-    label:
-      en_US: URL/Host/DSN
-      zh_Hans: URL/Host/DSN
-    human_description:
-      en_US: Please input depending on DB type, visit https://vanna.ai/docs/ for more specification
-      zh_Hans: 请根据数据库类型，填入对应值，详情参考https://vanna.ai/docs/
-    form: form
-  - name: db_name
-    type: string
-    required: false
-    label:
-      en_US: DB name
-      zh_Hans: 数据库名
-    human_description:
-      en_US: Database name
-      zh_Hans: 数据库名
-    form: form
-  - name: username
-    type: string
-    required: false
-    label:
-      en_US: Username
-      zh_Hans: 用户名
-    human_description:
-      en_US: Username
-      zh_Hans: 用户名
-    form: form
-  - name: password
-    type: secret-input
-    required: false
-    label:
-      en_US: Password
-      zh_Hans: 密码
-    human_description:
-      en_US: Password
-      zh_Hans: 密码
-    form: form
-  - name: port
-    type: number
-    required: false
-    label:
-      en_US: Port
-      zh_Hans: 端口
-    human_description:
-      en_US: Port
-      zh_Hans: 端口
-    form: form
-  - name: ddl
-    type: string
-    required: false
-    label:
-      en_US: Training DDL
-      zh_Hans: 训练DDL
-    human_description:
-      en_US: DDL statements for training data
-      zh_Hans: 用于训练RAG Model的建表语句
-    form: llm
-  - name: question
-    type: string
-    required: false
-    label:
-      en_US: Training Question
-      zh_Hans: 训练问题
-    human_description:
-      en_US: Question-SQL Pairs
-      zh_Hans: Question-SQL中的问题
-    form: llm
-  - name: sql
-    type: string
-    required: false
-    label:
-      en_US: Training SQL
-      zh_Hans: 训练SQL
-    human_description:
-      en_US: SQL queries to your training data
-      zh_Hans: 用于训练RAG Model的SQL语句
-    form: llm
-  - name: memos
-    type: string
-    required: false
-    label:
-      en_US: Training Memos
-      zh_Hans: 训练说明
-    human_description:
-      en_US: Sometimes you may want to add documentation about your business terminology or definitions
-      zh_Hans: 添加更多关于数据库的业务说明
-    form: llm
-  - name: enable_training
-    type: boolean
-    required: false
-    default: false
-    label:
-      en_US: Training Data
-      zh_Hans: 训练数据
-    human_description:
-      en_US: You only need to train once. Do not train again unless you want to add more training data
-      zh_Hans: 训练数据无更新时，训练一次即可
-    form: form
-  - name: reset_training_data
-    type: boolean
-    required: false
-    default: false
-    label:
-      en_US: Reset Training Data
-      zh_Hans: 重置训练数据
-    human_description:
-      en_US: Remove all training data in the current RAG Model
-      zh_Hans: 删除当前RAG Model中的所有训练数据
-    form: form
-  - name: training_metadata
-    type: boolean
-    required: false
-    default: false
-    label:
-      en_US: Training Metadata
-      zh_Hans: 训练元数据
-    human_description:
-      en_US: If enabled, it will attempt to train on the metadata of that database
-      zh_Hans: 是否自动从数据库获取元数据来训练
-    form: form
-  - name: allow_llm_to_see_data
-    type: boolean
-    required: false
-    default: false
-    label:
-      en_US: Whether to allow the LLM to see the data
-      zh_Hans: 是否允许LLM查看数据
-    human_description:
-      en_US: Whether to allow the LLM to see the data
-      zh_Hans: 是否允许LLM查看数据
-    form: form
@@ -1,46 +0,0 @@
-import re
-from typing import Any
-from urllib.parse import urlparse
-
-from core.tools.errors import ToolProviderCredentialValidationError
-from core.tools.provider.builtin.vanna.tools.vanna import VannaTool
-from core.tools.provider.builtin_tool_provider import BuiltinToolProviderController
-
-
-class VannaProvider(BuiltinToolProviderController):
-    def _get_protocol_and_main_domain(self, url):
-        parsed_url = urlparse(url)
-        protocol = parsed_url.scheme
-        hostname = parsed_url.hostname
-        port = f":{parsed_url.port}" if parsed_url.port else ""
-
-        # Check if the hostname is an IP address
-        is_ip = re.match(r"^\d{1,3}(\.\d{1,3}){3}$", hostname) is not None
-
-        # Return the full hostname (with port if present) for IP addresses, otherwise return the main domain
-        main_domain = f"{hostname}{port}" if is_ip else ".".join(hostname.split(".")[-2:]) + port
-        return f"{protocol}://{main_domain}"
-
-    def _validate_credentials(self, credentials: dict[str, Any]) -> None:
-        base_url = credentials.get("base_url")
-        if not base_url:
-            base_url = "https://ask.vanna.ai/rpc"
-        else:
-            base_url = base_url.removesuffix("/")
-        credentials["base_url"] = base_url
-        try:
-            VannaTool().fork_tool_runtime(
-                runtime={
-                    "credentials": credentials,
-                }
-            ).invoke(
-                user_id="",
-                tool_parameters={
-                    "model": "chinook",
-                    "db_type": "SQLite",
-                    "url": f"{self._get_protocol_and_main_domain(credentials['base_url'])}/Chinook.sqlite",
-                    "query": "What are the top 10 customers by sales?",
-                },
-            )
-        except Exception as e:
-            raise ToolProviderCredentialValidationError(str(e))
@@ -1,35 +0,0 @@
-identity:
-  author: QCTC
-  name: vanna
-  label:
-    en_US: Vanna.AI
-    zh_Hans: Vanna.AI
-  description:
-    en_US: The fastest way to get actionable insights from your database just by asking questions.
-    zh_Hans: 一个基于大模型和RAG的Text2SQL工具。
-  icon: icon.png
-  tags:
-    - utilities
-    - productivity
-credentials_for_provider:
-  api_key:
-    type: secret-input
-    required: true
-    label:
-      en_US: API key
-      zh_Hans: API key
-    placeholder:
-      en_US: Please input your API key
-      zh_Hans: 请输入你的 API key
-      pt_BR: Please input your API key
-    help:
-      en_US: Get your API key from Vanna.AI
-      zh_Hans: 从 Vanna.AI 获取你的 API key
-    url: https://vanna.ai/account/profile
-  base_url:
-    type: text-input
-    required: false
-    label:
-      en_US: Vanna.AI Endpoint Base URL
-    placeholder:
-      en_US: https://ask.vanna.ai/rpc
@@ -195,14 +195,14 @@ class WorkflowTool(Tool):
            if isinstance(value, list):
                for item in value:
                    if isinstance(item, dict) and item.get("dify_model_identity") == FILE_MODEL_IDENTITY:
-                        item = self._update_file_mapping(item)
+                        item["tool_file_id"] = item.get("related_id")
                        file = build_from_mapping(
                            mapping=item,
                            tenant_id=str(cast(Tool.Runtime, self.runtime).tenant_id),
                        )
                        files.append(file)
            elif isinstance(value, dict) and value.get("dify_model_identity") == FILE_MODEL_IDENTITY:
-                value = self._update_file_mapping(value)
+                value["tool_file_id"] = value.get("related_id")
                file = build_from_mapping(
                    mapping=value,
                    tenant_id=str(cast(Tool.Runtime, self.runtime).tenant_id),
@@ -211,11 +211,3 @@ class WorkflowTool(Tool):

            result[key] = value
        return result, files
-
-    def _update_file_mapping(self, file_dict: dict) -> dict:
-        transfer_method = FileTransferMethod.value_of(file_dict.get("transfer_method"))
-        if transfer_method == FileTransferMethod.TOOL_FILE:
-            file_dict["tool_file_id"] = file_dict.get("related_id")
-        elif transfer_method == FileTransferMethod.LOCAL_FILE:
-            file_dict["upload_file_id"] = file_dict.get("related_id")
-        return file_dict
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
-LAN-	4b938ab18d	chore: Bump version Signed-off-by: -LAN- <[email protected]>	2025-05-30 16:25:40 +08:00
-LAN-	88356de923	fix: Refactor web reader to use readabilipy (#19789 ) Signed-off-by: -LAN- <[email protected]>	2025-05-30 16:23:17 +08:00
-LAN-andGitHub	5f09900dca	chore(api): Upgrade dependencies (#19736 ) Signed-off-by: -LAN- <[email protected]>	2025-05-15 14:47:15 +08:00
-LAN-	9ac99abf20	docs(CHANGELOG): Update CHANGELOG Signed-off-by: -LAN- <[email protected]>	2025-05-14 18:03:05 +08:00
-LAN-	32588f562e	feat(model): fix and re-add gpt-4.1. Signed-off-by: -LAN- <[email protected]>	2025-05-14 18:02:32 +08:00
JoelandGitHub	36f8bd3f1a	chore: frontend third-part package security issue (#19655 )	2025-05-14 14:08:05 +08:00
-LAN-	c919074e06	docs(CHANGELOG.md): Update CHANGELOG.md Signed-off-by: -LAN- <[email protected]>	2025-05-13 10:31:40 +08:00
kelvintsim GitHub crazywoola Bowen Liang	88cd9aedb7	add gunicorn keepalive setting (#19537 ) Co-authored-by: crazywoola <[email protected]> Co-authored-by: Bowen Liang <[email protected]>	2025-05-13 10:28:13 +08:00
-LAN-	16a4f77fb4	fix(config): Allow DB_EXTRAS to set search_path via options Signed-off-by: -LAN- <[email protected]>	2025-05-13 10:19:08 +08:00
-LAN-andGitHub	3401c52665	chore(pyproject.toml): Upgrade huggingface-hub, transformers and resend (#19563 ) Signed-off-by: -LAN- <[email protected]>	2025-05-12 23:21:57 +08:00
Zixuan ChengandGitHub	4fa3d78ed8	Revert "feat : add GPT4.1 in the model providers" (#19002 )	2025-04-28 18:15:24 +08:00
-LAN-	5f7f851b17	fix: Refines None checks in result transformation Simplifies the code by replacing type checks for None with direct comparisons, improving readability and consistency in handling None values during output validation. Signed-off-by: -LAN- <[email protected]>	2025-04-28 15:40:14 +08:00
-LAN-	559ab46ee1	fix: Removes redundant token calculations and updates dependencies Eliminates unnecessary pre-calculation of token limits and recalculation of max tokens across multiple app runners, simplifying the logic for prompt handling. Updates tiktoken library from version 0.8.0 to 0.9.0 for improved tokenization performance. Increases default token limit in TokenBufferMemory to accommodate larger prompt messages. These changes streamline the token management process and leverage the latest improvements in the tiktoken library. Fixes potential token overflow issues and prepares the system for handling larger inputs more efficiently. Relates to internal optimization tasks. Signed-off-by: -LAN- <[email protected]>	2025-04-28 15:39:12 +08:00
-LAN-	df98223c8c	chore: Updates to version 0.15.7 with new model support Adds support for GPT-4.1 and Amazon Bedrock DeepSeek-R1 models. Fixes issues with app creation from template categories and DSL version checks. Updates version numbers in configuration files and Docker setup to 0.15.7 for consistency. Addresses issues #18807, #18868, #18872, #18878, and #18912. Signed-off-by: -LAN- <[email protected]>	2025-04-28 14:19:07 +08:00
Zixuan ChengandGitHub	144f9507f8	feat : add GPT4.1 in the model providers (#18912 )	2025-04-27 19:31:20 +08:00
kelvintsimandGitHub	2e097a1ac0	add bedrock deepseek-r1 (#18908 )	2025-04-27 19:30:42 +08:00
NFishandGitHub	9f7d8a981f	Patch: hotfix/create from template category (#18807 ) (#18868 )	2025-04-27 14:47:18 +08:00
zxhlyhandGitHub	40b31bafd5	fix: check dsl version when create app from explore template (#18872 ) (#18878 )	2025-04-27 14:21:45 +08:00
-LAN-	d38a2c95fb	docs(CHANGELOG): Update CHANGELOG.md Signed-off-by: -LAN- <[email protected]>	2025-04-25 18:31:08 +08:00
-LAN-	7d18e2a0ef	feat(app_dsl_service): Refines version compatibility logic Updates logic to handle various version comparisons, ensuring more precise status returns based on version differences. Improves handling of older and newer versions to prevent mismatches and ensure appropriate compatibility status. Signed-off-by: -LAN- <[email protected]>	2025-04-25 18:27:31 +08:00
kelvintsimandGitHub	024f242251	add bedrock claude-sonnet-3.7 (#18788 )	2025-04-25 17:35:12 +08:00
-LAN-	bfdce78ca5	chore(*): Bump up to 0.15.6 Signed-off-by: -LAN- <[email protected]>	2025-04-23 14:06:46 +08:00
-LAN-	00c2258352	CHANGELOG): Adds initial changelog for version 0.15.6 Signed-off-by: -LAN- <[email protected]>	2025-04-23 13:55:33 +08:00
JoelandGitHub	a1b3d41712	fix: clickjacking (#18552 )	2025-04-22 17:08:52 +08:00
kautsar_masuara GitHub achmad-kautsar	b26e20fe34	fix: fix vertex gemini 2.0 flash 001 schema (#18405 ) Co-authored-by: achmad-kautsar <[email protected]>	2025-04-19 22:04:13 +08:00
NFishandGitHub	161ff432f1	fix: update reset password token when email code verify success (#18362 )	2025-04-18 17:15:15 +08:00
Xiyuan ChenandGitHub	99a9def623	fix: reset_password security issue (#18366 )	2025-04-18 05:04:44 -04:00
Alexi.FandGitHub	fe1846c437	fix: change gemini-2.0-flash to validate google api #17082 (#17115 )	2025-03-30 13:04:12 +08:00
-LAN-	8e75eb5c63	fix: update version to 0.15.5 in packaging and docker-compose files Sgned-off-by: -LAN- <[email protected]>	2025-03-24 16:47:06 +08:00
-LAN-	970508fcb6	fix: update GitHub Actions workflow to trigger on tags Signed-off-by: -LAN- <[email protected]>	2025-03-24 16:45:29 +08:00
NFish	9283a5414f	fix: update yarn.lock	2025-03-24 16:41:07 +08:00
-LAN-	2a2a0e9be9	fix: update DifySandbox image version to 0.2.11 in docker-compose files Sgned-off-by: -LAN- <[email protected]>	2025-03-24 15:37:55 +08:00
JoelandGitHub	061a765b7d	fix: sanitizer svg to avoid xss (#16608 )	2025-03-24 14:48:40 +08:00
-LAN-	acd7fead87	feat: remove Vanna provider and associated assets from the project Signed-off-by: -LAN- <[email protected]>	2025-03-24 14:34:03 +08:00
NFish	bbb080d5b2	fix: update chatbot help doc link on the create app form	2025-03-24 11:28:35 +08:00
NFish	c01d8a70f3	fix: upgrade nextjs to v14.2.25. a security patch for CVE-2025-29927.	2025-03-24 10:32:18 +08:00
-LAN-	1ca15989e0	chore: update version to 0.15.4 in configuration and docker files Signed-off-by: -LAN- <[email protected]>	2025-03-21 16:39:06 +08:00
-LAN-	8b5a3a9424	Merge branch 'release/0.15.4' of github.com:langgenius/dify into release/0.15.4	2025-03-21 16:31:06 +08:00
-LAN-	42ddcf1edd	chore: remove 0.15.3 branch config in the build action Signed-off-by: -LAN- <[email protected]>	2025-03-21 16:30:33 +08:00
JoelandGitHub	21561df10f	fix: xss in render svg (#16437 )	2025-03-21 15:24:58 +08:00
crazywoola	0e33a3aa5f	chore: add ci	2025-02-19 14:34:36 +08:00
Hash Brownandcrazywoola	d3895bcd6b	revert	2025-02-19 14:32:28 +08:00
Hash Brownandcrazywoola	eeb390650b	fix: build failed	2025-02-19 14:32:28 +08:00