Swarms — оркестрация мультиагентных систем

★ 7.1K

kyegomez/swarms is an open-source Python framework for multi-agent orchestration that connects multiple AI agents into a coordinated workflow with explicit message-passing between them. Reach for it when a single agent isn't enough: you need to split a long research report into gathering, fact-checking, and editing stages; route incoming requests across several specialized roles; or run an independent verification pass on a complex calculation. The framework ships prebuilt interaction patterns — sequential chains, concurrent fan-out, hierarchical supervisor-worker setups — along with shared memory across agents, tool integrations, and Model Context Protocol (MCP) support, plus backward compatibility with other agent frameworks. Install via pip, uv, or poetry; model API keys are set through environment variables. The framework orchestrates agent dialogue and assembles the final output from the topology you define — it is a construction kit for agent teams, not a ready-made product with a UI. Compared to LangGraph, it favors predefined team topologies over manual state-graph assembly; compared to CrewAI, it offers a wider range of topologies and protocol integrations. Requires API ke