Implementation:CrewAIInc CrewAI Bedrock KB Retriever Tool

Knowledge Sources	CrewAI
Domains	Cloud_Integration, AWS_Bedrock, RAG, Knowledge_Retrieval
Last Updated	2026-02-11 00:00 GMT

Overview

Concrete tool for retrieving information from Amazon Bedrock Knowledge Bases provided by CrewAI.

Description

The BedrockKBRetrieverTool class extends BaseTool to perform semantic search against Amazon Bedrock Knowledge Bases using the boto3 Bedrock Agent Runtime client. It accepts a natural language query and returns structured JSON results with content, metadata, relevance scores, and source provenance. The tool supports configurable number_of_results (default: 5), custom retrieval_configuration for vector search settings, guardrail_configuration for content safety, and pagination via next_token. The _process_retrieval_result() method handles multiple data source types including S3, Confluence, Salesforce, SharePoint, Web, Custom Documents, Kendra Documents, and SQL locations, extracting the appropriate URI or identifier for each. Parameter validation enforces constraints on knowledge base ID format (alphanumeric, max 10 characters) and optional token format. Configuration values can be provided via constructor or BEDROCK_KB_ID environment variable.

Usage

Import and instantiate BedrockKBRetrieverTool when CrewAI agents need to ground their responses in enterprise knowledge stored in Amazon Bedrock Knowledge Bases. This enables agents to perform semantic search across documents, wikis, databases, and other data sources indexed by Bedrock.

Code Reference

Source Location

• Repository: CrewAI
• File: lib/crewai-tools/src/crewai_tools/aws/bedrock/knowledge_base/retriever_tool.py
• Lines: 1-269

Signature

class BedrockKBRetrieverTool(BaseTool):
    def __init__(
        self,
        knowledge_base_id: str | None = None,
        number_of_results: int | None = 5,
        retrieval_configuration: dict[str, Any] | None = None,
        guardrail_configuration: dict[str, Any] | None = None,
        next_token: str | None = None,
        **kwargs,
    ):

Import

from crewai_tools.aws.bedrock.knowledge_base import BedrockKBRetrieverTool

I/O Contract

Inputs (Constructor)

Name	Type	Required	Description
knowledge_base_id	None	No	Unique ID of the Bedrock Knowledge Base (alphanumeric, max 10 chars); falls back to BEDROCK_KB_ID env var
number_of_results	None	No	Maximum number of results to return (default: 5)
retrieval_configuration	None	No	Custom retrieval configuration for vector search settings
guardrail_configuration	None	No	Guardrail settings for content safety
next_token	None	No	Token for paginating through results

Inputs (_run)

Name	Type	Required	Description
query	str	Yes	Natural language query to search the knowledge base

Outputs

Name	Type	Description
return	str	JSON string containing results array with content, content_type, source_type, source_uri, optional score and metadata; or a message if no results found; may include nextToken and guardrailAction fields

Usage Examples

Basic Usage

from crewai_tools.aws.bedrock.knowledge_base import BedrockKBRetrieverTool

tool = BedrockKBRetrieverTool(
    knowledge_base_id="ABC1234567",
    number_of_results=10,
)

# Use with a CrewAI agent
agent = Agent(
    role="research_analyst",
    tools=[tool],
)

With Custom Retrieval Configuration

from crewai_tools.aws.bedrock.knowledge_base import BedrockKBRetrieverTool

tool = BedrockKBRetrieverTool(
    knowledge_base_id="ABC1234567",
    retrieval_configuration={
        "vectorSearchConfiguration": {
            "numberOfResults": 10,
            "overrideSearchType": "HYBRID",
        }
    },
    guardrail_configuration={
        "guardrailId": "my-guardrail",
        "guardrailVersion": "1",
    },
)

result = tool._run(query="What are the quarterly revenue figures?")

Related Pages

• Principle:CrewAIInc_CrewAI_Built_In_Tool_Selection