Open Multi Agent Orchestration

This is for building multi-agent AI systems where different agents with specialized roles collaborate on complex tasks. You get task dependency graphs (DAG scheduling), parallel execution, and shared memory, all running in-process without subprocess overhead. The framework handles both single-agent workflows and full teams where you can mix Claude and GPT models in the same pipeline. Particularly good when you need explicit control over who does what and in what order, like having an architect agent design an API spec, then blocking a developer agent until that's done, while a reviewer agent runs in parallel. Includes built-in tools for file operations and bash, plus you can define custom tools with Zod schemas.