Agent Squad is a framework for orchestrating multiple AI agents and managing complex conversations across Python, TypeScript, and Swift runtimes.
The tool solves the problem of coordinating specialized agents to handle diverse user queries by routing each request to the most suitable agent while maintaining conversation context. It uses an intelligent classifier that analyzes user input and conversation history to select the best agent, which then processes the request and may invoke tools as needed. The orchestrator manages the exchange and preserves context across agent interactions.
Developers should choose this tool if they need to deploy multi-agent systems across cloud and on-device environments. Python and TypeScript implementations suit cloud deployments on AWS Lambda or containers, while the Swift runtime enables on-device orchestration for iOS and macOS applications. The framework provides pre-built agents for services like Bedrock, Anthropic, OpenAI, and AWS Lex, along with extensibility points for custom agents, classifiers, storage backends, and retrievers. The Swift runtime distinguishes itself by supporting entirely local execution with classifier routing, native and Model Context Protocol tools, realtime voice capabilities, and local-first chat storage.
The project maintains active development across all three runtimes with regular updates to support new AI providers and capabilities. The codebase demonstrates careful attention to extensibility through well-defined integration points rather than monolithic implementations. Documentation is comprehensive, covering each runtime separately while emphasizing the shared orchestration model. The project recently transitioned to a new home and maintains backward compatibility considerations for users migrating from its previous location.
