Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
feat: improvements on QdrantVectorSearchTool
Daniel Barreto2ee27efca7a8
diff --git a/lib/crewai-tools/src/crewai_tools/tools/qdrant_vector_search_tool/qdrant_search_tool.py b/lib/crewai-tools/src/crewai_tools/tools/qdrant_vector_search_tool/qdrant_search_tool.pyindex c1a88114e..adcb733a5 100644--- a/lib/crewai-tools/src/crewai_tools/tools/qdrant_vector_search_tool/qdrant_search_tool.py+++ b/lib/crewai-tools/src/crewai_tools/tools/qdrant_vector_search_tool/qdrant_search_tool.py@@ -1,80 +1,42 @@-from collections.abc import Callable+from __future__ import annotations++import importlibimport jsonimport os+from collections.abc import Callablefrom typing import Any--try:- from qdrant_client import QdrantClient- from qdrant_client.http.models import FieldCondition, Filter, MatchValue-- QDRANT_AVAILABLE = True-except ImportError:- QDRANT_AVAILABLE = False- QdrantClient = Any # type: ignore[assignment,misc] # type placeholder- Filter = Any # type: ignore[assignment,misc]- FieldCondition = Any # type: ignore[assignment,misc]- MatchValue = Any # type: ignore[assignment,misc]-from crewai.tools import BaseTool, EnvVar-from pydantic import BaseModel, ConfigDict, Field+from pydantic import BaseModel, ConfigDict, Field, model_validator+from pydantic.types import ImportStringclass QdrantToolSchema(BaseModel):- """Input for QdrantTool."""+ query: str = Field(..., description="Query to search in Qdrant DB.")+ filter_by: str | None = None+ filter_value: str | None = None- query: str = Field(- ...,- description="The query to search retrieve relevant information from the Qdrant database. Pass only the query, not the question.",- )- filter_by: str | None = Field(- default=None,- description="Filter by properties. Pass only the properties, not the question.",- )- filter_value: str | None = Field(- default=None,- description="Filter by value. Pass only the value, not the question.",- )+class QdrantConfig(BaseModel):+ """All Qdrant connection and search settings."""-class QdrantVectorSearchTool(BaseTool):- """Tool to query and filter results from a Qdrant database.+ qdrant_url: str+ qdrant_api_key: str | None = None+ collection_name: str+ limit: int = 3+ score_threshold: float = 0.35+ filter_conditions: list[tuple[str, Any]] = Field(default_factory=list)- This tool enables vector similarity search on internal documents stored in Qdrant,- with optional filtering capabilities.- Attributes:- client: Configured QdrantClient instance- collection_name: Name of the Qdrant collection to search- limit: Maximum number of results to return- score_threshold: Minimum similarity score threshold- qdrant_url: Qdrant server URL- qdrant_api_key: Authentication key for Qdrant- """+class QdrantVectorSearchTool(BaseTool):+ """Vector search tool for Qdrant."""model_config = ConfigDict(arbitrary_types_allowed=True)- client: QdrantClient = None # type: ignore[assignment]++ # --- Metadata ---name: str = "QdrantVectorSearchTool"- description: str = "A tool to search the Qdrant database for relevant information on internal documents."+ description: str = "Search Qdrant vector DB for relevant documents."args_schema: type[BaseModel] = QdrantToolSchema- query: str | None = None- filter_by: str | None = None- filter_value: str | None = None- collection_name: str | None = None- limit: int | None = Field(default=3)- score_threshold: float = Field(default=0.35)- qdrant_url: str = Field(- ...,- description="The URL of the Qdrant server",- )- qdrant_api_key: str | None = Field(- default=None,- description="The API key for the Qdrant server",- )- custom_embedding_fn: Callable | None = Field(- default=None,- description="A custom embedding function to use for vectorization. If not provided, the default model will be used.",- )package_dependencies: list[str] = Field(default_factory=lambda: ["qdrant-client"])env_vars: list[EnvVar] = Field(default_factory=lambda: [@@ -83,107 +45,81 @@ class QdrantVectorSearchTool(BaseTool):)])-- def __init__(self, **kwargs):- super().__init__(**kwargs)- if QDRANT_AVAILABLE:- self.client = QdrantClient(- url=self.qdrant_url,- api_key=self.qdrant_api_key if self.qdrant_api_key else None,+ qdrant_config: QdrantConfig+ qdrant_package: ImportString[Any] = Field(+ default="qdrant_client",+ description="Base package path for Qdrant. Will dynamically import client and models.",+ )+ custom_embedding_fn: ImportString[Callable[[str], list[float]]] | None = Field(+ default=None,+ description="Optional embedding function or import path.",+ )+ client: Any | None = None++ @model_validator(mode="after")+ def _setup_qdrant(self) -> QdrantVectorSearchTool:+ # Import the qdrant_package if it's a string+ if isinstance(self.qdrant_package, str):+ self.qdrant_package = importlib.import_module(self.qdrant_package)++ if not self.client:+ self.client = self.qdrant_package.QdrantClient(+ url=self.qdrant_config.qdrant_url,+ api_key=self.qdrant_config.qdrant_api_key or None,)- else:- import click-- if click.confirm(- "The 'qdrant-client' package is required to use the QdrantVectorSearchTool. "- "Would you like to install it?"- ):- import subprocess-- subprocess.run(["uv", "add", "qdrant-client"], check=True) # noqa: S607- else:- raise ImportError(- "The 'qdrant-client' package is required to use the QdrantVectorSearchTool. "- "Please install it with: uv add qdrant-client"- )+ return selfdef _run(self,query: str,filter_by: str | None = None,- filter_value: str | None = None,+ filter_value: Any | None = None,) -> str:- """Execute vector similarity search on Qdrant.-- Args:- query: Search query to vectorize and match- filter_by: Optional metadata field to filter on- filter_value: Optional value to filter by-- Returns:- JSON string containing search results with metadata and scores-- Raises:- ImportError: If qdrant-client is not installed- ValueError: If Qdrant credentials are missing- """- if not self.qdrant_url:- raise ValueError("QDRANT_URL is not set")-- # Create filter if filter parameters are provided- search_filter = None- if filter_by and filter_value:- search_filter = Filter(+ """Perform vector similarity search."""+ filter_ = self.qdrant_package.http.models.Filter+ field_condition = self.qdrant_package.http.models.FieldCondition+ match_value = self.qdrant_package.http.models.MatchValue+ conditions = self.qdrant_config.filter_conditions.copy()+ if filter_by and filter_value is not None:+ conditions.append((filter_by, filter_value))++ search_filter = (+ filter_(must=[- FieldCondition(key=filter_by, match=MatchValue(value=filter_value))+ field_condition(key=k, match=match_value(value=v))+ for k, v in conditions])-- # Search in Qdrant using the built-in query method+ if conditions+ else None+ )query_vector = (- self._vectorize_query(query, embedding_model="text-embedding-3-large")- if not self.custom_embedding_fn- else self.custom_embedding_fn(query)+ self.custom_embedding_fn(query)+ if self.custom_embedding_fn+ else (+ lambda: __import__("openai")+ .Client(api_key=os.getenv("OPENAI_API_KEY"))+ .embeddings.create(input=[query], model="text-embedding-3-large")+ .data[0]+ .embedding+ )())- search_results = self.client.query_points(- collection_name=self.collection_name, # type: ignore[arg-type]+ results = self.client.query_points(+ collection_name=self.qdrant_config.collection_name,query=query_vector,query_filter=search_filter,- limit=self.limit, # type: ignore[arg-type]- score_threshold=self.score_threshold,+ limit=self.qdrant_config.limit,+ score_threshold=self.qdrant_config.score_threshold,)- # Format results similar to storage implementation- results = []- # Extract the list of ScoredPoint objects from the tuple- for point in search_results:- result = {- "metadata": point[1][0].payload.get("metadata", {}),- "context": point[1][0].payload.get("text", ""),- "distance": point[1][0].score,- }- results.append(result)-- return json.dumps(results, indent=2)-- def _vectorize_query(self, query: str, embedding_model: str) -> list[float]:- """Default vectorization function with openai.-- Args:- query (str): The query to vectorize- embedding_model (str): The embedding model to use-- Returns:- list[float]: The vectorized query- """- import openai-- client = openai.Client(api_key=os.getenv("OPENAI_API_KEY"))- return (- client.embeddings.create(- input=[query],- model=embedding_model,- )- .data[0]- .embedding+ return json.dumps(+ [+ {+ "distance": p.score,+ "metadata": p.payload.get("metadata", {}) if p.payload else {},+ "context": p.payload.get("text", "") if p.payload else {},+ }+ for p in results.points+ ],+ indent=2,)