langroid

Langroid is an intuitive, lightweight, extensible and principled Python framework for building LLM-powered applications, from researchers at CMU and UW-Madison. You set up agents, equip them with optional components (an LLM, a vector store and tools or functions), assign them tasks, and let them solve a problem by exchanging messages. The multi-agent paradigm is inspired by the Actor model. The README says it does not use LangChain or any other LLM framework and works with practically any LLM.

The core abstractions are the Agent, which encapsulates conversation state and optionally a vector store and tools and acts as a message transformer with responder methods for the LLM, the agent and the user, and the Task, which wraps an agent, gives it instructions and orchestrates interactions through hierarchical, recursive task delegation. Listed items include MCP support so any agent can use MCP servers, a Claude Code plugin with built-in patterns, and extras installed with pip (for example `pip install "langroid[doc-chat]"`, `[db]` and `[hf-embeddings]`). It requires Python 3.11+, all of Langroid is available through `import langroid as lr`, and the code is MIT licensed.