Skip to content

Bedrock Agency Swarm

A framework for building multi-agent systems using Amazon Bedrock.

Features

  • Multi-Agent Communication: Coordinate multiple specialized agents to solve complex tasks
  • Coordinator-Based Architecture: Smart task planning and delegation
  • Event Tracing: Comprehensive tracing of all agent interactions and tool executions
  • Tool Integration: Easy integration of custom tools and capabilities
  • Flexible Memory: Configurable memory systems for agent context

Documentation

  • Architecture: Learn about the coordinator-based design and communication flow
  • Examples: See practical examples of using the framework
  • API Reference: Detailed API documentation

Quick Start

  1. Install the package: 

    pip install bedrock-agency-swarm

  2. Set up your AWS credentials: 

    export AWS_PROFILE=your-profile
export AWS_REGION=your-region

  3. Create a simple agency: 

    from bedrock_swarm.agency import Agency
from bedrock_swarm.agents import BedrockAgent
from bedrock_swarm.tools import CalculatorTool

# Create a specialist agent
calculator = BedrockAgent(
    name="calculator",
    model_id="us.anthropic.claude-3-5-sonnet-20241022-v2:0",
    tools=[CalculatorTool()],
    system_prompt="You are a mathematical specialist."
)

# Create agency with the specialist
agency = Agency(specialists=[calculator])

# Process a request
response = agency.process_request("What is 15 * 7?")
print(response)

Contributing

Contributions are welcome! Please read our Contributing Guide for details on our code of conduct and the process for submitting pull requests.

graph TD
    A[User Query] --> B[Agency]
    B --> C[Coordinator Agent]
    C --> D[Calculator Agent]
    C --> E[Time Expert Agent]
    C --> F[Other Specialists...]
    D --> G[Tools]
    E --> G
    F --> G
    G --> C
    C --> B
    B --> H[User Response]

Key Features

  • Multi-Agent Architecture: Create specialized agents that work together to solve complex tasks
  • Event-Driven Communication: Track and monitor all agent interactions and tool executions
  • Built-in Tool System: Easily extend agents with custom tools and capabilities
  • Memory Management: Share information between agents with flexible memory systems
  • AWS Bedrock Integration: Leverage powerful language models through Amazon Bedrock

Core Concepts

  1. Agency: The main orchestrator that manages communication between agents
  2. Agents: Specialized AI assistants with specific capabilities
  3. Tools: Functions that agents can use to perform specific tasks
  4. Threads: Conversation flows between users and agents
  5. Memory: System for storing and sharing information between agents
  6. Events: Detailed tracking of all system interactions

Next Steps