Agent Squad

One framework. Any LLM. Any platform. Coordinate multiple AI agents with intelligent routing, pluggable storage, and streaming support.

Available runtimes

Python / TypeScript

Full-featured orchestration for server and cloud: Bedrock, Anthropic, and OpenAI classifiers, streaming responses, Lambda/DynamoDB/SQL storage, retrievers, and a rich cookbook.

Docs → npm PyPI

Swift — Apple platforms

Protocol-driven runtime for iOS 16+ / macOS 14+: on-device agents, MCP tools, realtime voice, rich Tool UIs, SwiftData storage, and built-in tracing.

Docs → GitHub

Why Agent Squad

Bring your own LLM

Swap classifiers and agents independently. Works with Bedrock, Anthropic, OpenAI, Ollama, or anything you wire up — no lock-in.

Extensible by design

Every interface is a protocol or abstract class. Custom agents, classifiers, storage backends, and retrievers all follow the same simple contract.

Multi-agent orchestration

Register multiple specialized agents and let the classifier route each turn to the right one — automatically, based on context and conversation history.

Single-agent simplicity

No need to run a fleet. Use a single agent with tools, a retriever, or a custom prompt — the orchestrator stays out of the way.