Implementation:CrewAIInc CrewAI RAG Tool

Knowledge Sources	CrewAI
Domains	Tools, RAG, Knowledge_Base
Last Updated	2026-02-11 00:00 GMT

Overview

RagTool is the foundational base class for all knowledge-base-backed search tools in CrewAI, implementing the core RAG (Retrieval-Augmented Generation) pipeline.

Description

RagTool extends BaseTool and provides the central RAG abstraction in the CrewAI tools ecosystem. It defines an abstract Adapter base class with query() and add() methods, and uses a placeholder adapter pattern: on initialization, a model_validator detects the _AdapterPlaceholder and replaces it with a CrewAIRagAdapter configured from RagToolConfig. The _parse_config() method normalizes configuration into provider-specific objects, supporting chromadb (default) and qdrant as vector database backends. A _validate_embedding_config() helper provides clear, provider-specific error messages when ProviderSpec union validation fails. The add() method delegates document ingestion to the adapter, accepting content items with various keyword parameters (data_type, path, url, etc.). The _run() method queries the adapter with configurable similarity threshold and result limit, returning formatted relevant content. The _create_provider_config() static method instantiates the appropriate provider config with optional embedding function injection.

Usage

Use this tool as a general-purpose knowledge base for agents, or extend it (as PDFSearchTool does) to create specialized search tools for specific data types. It supports configurable vector database backends and embedding models.

Code Reference

Source Location

• Repository: CrewAI
• File: lib/crewai-tools/src/crewai_tools/tools/rag/rag_tool.py
• Lines: 1-263

Signature

class Adapter(BaseModel, ABC):
    def query(self, question: str, similarity_threshold: float | None = None,
              limit: int | None = None) -> str
    def add(self, *args: ContentItem, **kwargs: Unpack[AddDocumentParams]) -> None

class RagTool(BaseTool):
    name: str = "Knowledge base"
    description: str = "A knowledge base that can be used to answer questions."
    summarize: bool = False
    similarity_threshold: float = 0.6
    limit: int = 5
    collection_name: str = "rag_tool_collection"
    adapter: Adapter = Field(default_factory=_AdapterPlaceholder)
    config: RagToolConfig = Field(default_factory=RagToolConfig)

    def add(self, *args: ContentItem, **kwargs: Unpack[AddDocumentParams]) -> None
    def _run(self, query: str, similarity_threshold: float | None = None,
             limit: int | None = None) -> str

Import

from crewai_tools import RagTool

I/O Contract

Inputs

Name	Type	Required	Description
query	str	Yes	Question to search the knowledge base with
similarity_threshold	float or None	No	Minimum similarity score for results (default 0.6)
limit	int or None	No	Maximum number of results to return (default 5)

Outputs

Name	Type	Description
_run() returns	str	Formatted string "Relevant Content:\n{results}" from the adapter query

Usage Examples

Basic Usage

from crewai_tools import RagTool

# Default ChromaDB backend
tool = RagTool()
tool.add("path/to/document.pdf", data_type="pdf_file")
result = tool._run(query="What are the main conclusions?")

# With Qdrant backend and custom embedding model
tool = RagTool(
    config={
        "vectordb": {"provider": "qdrant", "config": {"url": "http://localhost:6333"}},
        "embedding_model": {"provider": "openai", "config": {"model": "text-embedding-3-large"}}
    },
    collection_name="my_collection",
    similarity_threshold=0.7,
    limit=10
)

Related Pages

• Principle:CrewAIInc_CrewAI_Built_In_Tool_Selection