Agent Squad
Flexible and powerful framework for managing multiple AI agents and handling complex conversations
Last verified:
What is Agent Squad?
Agent Squad is an open-source framework for building, orchestrating, and running multiple specialized AI agents that collaboratively handle complex conversational and task-oriented workflows. It provides an Orchestrator and Agent abstractions, built-in agent types (LLM agents, Bedrock/OpenAI agents, Lambda, and connectors), and tools for routing, classification, and persistent storage to coordinate multi-agent conversations. The project targets developers and teams building advanced chatbot systems, multi-agent research prototypes, and production integrations—especially those using AWS Bedrock or wanting flexible connectors to different LLM providers. Its design emphasizes extensibility, offering clear extension points for custom agents, storage backends, and deployment patterns so engineering teams can adapt it to serverless, containerized, or on-prem environments.
Agent Squad pricing
Pricing model: Freemium
Agent Squad is distributed as an open-source project with no licensing cost for the framework itself; there is no vendor-hosted pricing or paid plans listed on the site. Usage costs depend on external services you connect (for example, AWS Bedrock, OpenAI, or hosted compute and storage) and any infrastructure you deploy for production (Lambda, EC2, DynamoDB), which are billed separately by their providers. There is no built-in paid tier or managed hosting sold by the project; all functionality is available from the repository and documentation.
Agent Squad pros
- Open-source framework under active development
- Provides an Orchestrator abstraction for multi-agent coordination
- Includes built-in agent types (Bedrock, OpenAI, Lambda, Lex)
- Supports classifier-based routing between agents
- Quickstart guides and examples for TypeScript and Python
- Pluggable storage backends, including in-memory and DynamoDB
- Agent Overlap Analysis to optimize agent responsibilities
- Designed to run with AWS Bedrock integrations out of the box
- Extensible API to implement custom agents and tools
- Ability to combine multiple LLM providers in one system
- Documentation includes step-by-step quickstarts
- Supports persistent conversation storage for stateful agents
- Community and GitHub issues/discussions for support
- Releases and versioning available through repository
- Works in serverless patterns such as AWS Lambda
Agent Squad cons
- Requires developer effort to integrate and configure providers
- Strong AWS orientation may need adaptation for non-AWS stacks
- Some advanced features assume access to proprietary Bedrock models
- Operational complexity when coordinating many agents in production
- Documentation can require reading multiple pages/repo files to piece together setup
- No hosted SaaS offering—self-hosting required
- Certain connectors and examples target specific SDK versions
- Scaling, monitoring, and observability must be implemented by the user
Frequently asked questions about Agent Squad
What is Agent Squad and who should use it?
Agent Squad is an open-source multi-agent orchestration framework for building systems where multiple specialized agents collaborate to answer queries or perform tasks; it’s aimed at developers, researchers, and engineering teams building complex conversational agents or multi-model integrations.
Which agent types are included out of the box?
The project includes multiple built-in agent implementations such as Bedrock LLM Agent, OpenAI agent examples, Lambda agent, Amazon Lex bot integration, and classifier components to route queries among agents.
How do I get started with Agent Squad?
The documentation provides quickstart guides and examples for TypeScript and Python showing how to install the framework, create an Orchestrator, register agents, and run a multi-agent conversation locally or with AWS Bedrock credentials.
Does Agent Squad require AWS Bedrock?
No—Agent Squad supports multiple providers and connectors, but many examples and integrations target AWS Bedrock; you can adapt it to other providers like OpenAI or custom LLMs by implementing or wiring the appropriate agent classes.
How is routing between agents handled?
Routing uses a classifier agent or routing logic within the Orchestrator to analyze incoming queries and select the best agent(s) to handle them; users can configure or replace the classifier to suit their routing criteria.
Can I persist conversations and agent state?
Yes—the framework supports pluggable storage backends (examples include in-memory and DynamoDB) so you can persist conversations and agent state for long-running or stateful interactions when you configure the storage provider.
Is there a hosted or managed version available?
No; Agent Squad is provided as an open-source repository with no hosted managed offering; deployment, hosting, and operational tooling must be provided by your team or cloud provider.
How do I add custom agents or tools?
The framework exposes extension points and base Agent/Orchestrator classes that you can subclass or implement to add custom agents, tool integrations, or new storage adapters; examples in the repo illustrate typical extension patterns.
What production considerations should I plan for?
You should plan for scaling orchestration, monitoring and observability, cost management for external LLMs, secure credential handling for provider APIs, and durable storage—Agent Squad provides the framework but not the managed ops components.
Where can I report issues or contribute?
The project accepts issues, pull requests, and discussion on its GitHub repository and associated discussions pages; follow contribution guidelines in the repo to submit fixes, feature requests, or documentation improvements.