Langroid

About

Langroid came out of CMU and UW-Madison research as an alternative to monolithic agent chains: each agent is an actor that transforms messages, and Task objects compose them into parent-child hierarchies with explicit delegation and addressing. The library ships RAG agents, structured extraction, SQL/table agents, and vector-store integrations (Qdrant, Chroma, Milvus, pgvector, and others), and since v0.53.0 any agent can consume MCP server tools through an adapter. It targets Python developers building multi-agent LLM applications, including production users such as Nullify, and is released under MIT with monthly releases.