Implementation:CrewAIInc CrewAI CrewAI RAG Adapter

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

Overview

Adapter implementation that integrates CrewAI's native RAG (Retrieval Augmented Generation) system with the tool framework, supporting multiple vector database backends.

Description

The CrewAIRagAdapter class implements the Adapter interface and serves as the bridge between CrewAI's RAG system and the tool framework. It provides a unified interface for adding diverse content types to vector databases and querying them semantically.

Key capabilities:

• Configurable backends: Supports multiple vector stores through RagConfigType, including Qdrant and ChromaDB, with special handling for Qdrant-specific configuration (vectors_config) through type-safe duck typing.
• Semantic querying: The query() method searches the knowledge base with configurable similarity threshold (default 0.6) and result limit (default 5).
• Comprehensive content ingestion: The add() method supports strings, file paths, URLs, websites, GitHub repos, YouTube videos, and directories with automatic data type detection.
• Recursive directory processing: Walks directories, skipping binary files (pyc, png, jpg, zip, etc.), hidden files/directories, and __pycache__ directories.
• Document chunking: Uses data-type-specific loaders and chunkers to split content into embeddable segments.
• Deterministic IDs: Generates unique chunk IDs using SHA256 hashing of doc_id, chunk index, and content.
• Metadata sanitization: Sanitizes metadata for ChromaDB compatibility before storage.

Usage

Use this adapter when you need to add knowledge from files, URLs, or directories to a vector database for semantic search by CrewAI agents. It is the default RAG adapter for CrewAI's knowledge management system.

Code Reference

Source Location

• Repository: CrewAI
• File: lib/crewai-tools/src/crewai_tools/adapters/crewai_rag_adapter.py
• Lines: 1-355

Signature

class CrewAIRagAdapter(Adapter):
    """Adapter that uses CrewAI's native RAG system."""

    collection_name: str = "default"
    summarize: bool = False
    similarity_threshold: float = 0.6
    limit: int = 5
    config: RagConfigType | None = None
    _client: BaseClient | None = PrivateAttr(default=None)

Import

from crewai_tools.adapters.crewai_rag_adapter import CrewAIRagAdapter

I/O Contract

Inputs (Constructor)

Name	Type	Required	Description
collection_name	str	No	Name of the vector collection (default: "default")
summarize	bool	No	Whether to summarize content (default: False)
similarity_threshold	float	No	Minimum similarity score for search results (default: 0.6)
limit	int	No	Maximum number of results to return (default: 5)
config	RagConfigType or None	No	Custom vector database configuration (default: None, uses default client)

query() Method

Name	Type	Required	Description
question	str	Yes	The question to search for
similarity_threshold	float or None	No	Override the instance-level threshold
limit	int or None	No	Override the instance-level result limit

add() Method

Name	Type	Required	Description
*args	ContentItem	No	Content items (strings, paths, or document dicts)
data_type	DataType or str	No	Explicit data type ("file", "pdf_file", "text", etc.)
path	str	No	Path to file or directory
file_path	str	No	Alias for path
metadata	dict	No	Additional metadata to attach to documents
url	str	No	URL to fetch content from
website	str	No	Website URL to scrape
github_url	str	No	GitHub repository URL
youtube_url	str	No	YouTube video URL
directory_path	str	No	Path to directory for recursive processing

Outputs

Name	Type	Description
query() return	str	Concatenated relevant content from the knowledge base, or "No relevant content found."
add() return	None	Documents are added to the vector store (no return value)

Usage Examples

Basic Usage

from crewai_tools.adapters.crewai_rag_adapter import CrewAIRagAdapter

# Create adapter with default ChromaDB backend
adapter = CrewAIRagAdapter(collection_name="my_knowledge")

# Add a PDF document
adapter.add("path/to/document.pdf", data_type="pdf_file")

# Add from a URL
adapter.add(url="https://example.com/article")

# Add an entire directory
adapter.add(directory_path="path/to/docs/")

# Query the knowledge base
result = adapter.query("What is the main topic?", similarity_threshold=0.7, limit=3)

With Qdrant Backend

from crewai_tools.adapters.crewai_rag_adapter import CrewAIRagAdapter

adapter = CrewAIRagAdapter(
    collection_name="qdrant_collection",
    config=qdrant_config,  # QdrantConfig with vectors_config
    similarity_threshold=0.7,
    limit=10,
)