Langroid

Langroid is an open-source Python framework designed to simplify the development of LLM-powered multi-agent systems. It provides a principled, intuitive abstraction for building agents that can collaborate, use tools, and interact with documents, databases, and APIs. Langroid is built with a "batteries included" philosophy, offering rich built-in support for RAG, function calling, and agent orchestration out of the box. It is well-suited for researchers and developers who want fine-grained control over agent behavior without sacrificing ease of use.

• Multi-Agent Collaboration: Define multiple agents with distinct roles and have them communicate via a message-passing architecture to solve complex tasks.
• Tool/Function Calling: Equip agents with tools (Python functions) that can be invoked during conversations, supporting both OpenAI-style function calling and custom tool definitions.
• Retrieval-Augmented Generation (RAG): Built-in support for document ingestion, chunking, embedding, and vector-store retrieval to ground agent responses in external knowledge.
• LLM Agnostic: Works with OpenAI, Azure OpenAI, Anthropic, local models via Ollama/llama.cpp, and other providers through a unified interface.
• Task Orchestration: Use the Task abstraction to wrap agents and compose them into pipelines or hierarchies, with automatic handling of turn-taking and termination conditions.
• Vector Store Integrations: Supports Qdrant, Chroma, Weaviate, LanceDB, and other vector databases for flexible RAG pipelines.
• Structured Outputs: Leverage Pydantic models for typed, validated agent inputs and outputs, making it easy to build reliable pipelines.
• Async & Streaming Support: Supports async execution and streaming responses for building responsive, production-grade applications.
• Extensible Architecture: Easily subclass ChatAgent or Task to customize behavior, add new tools, or integrate with external services.
• Rich Documentation & Examples: Comprehensive docs and a large library of example scripts covering common patterns like two-agent chat, document QA, SQL agents, and more.

To get started, install via pip install langroid, set your LLM API keys, and follow the quickstart guide in the documentation to build your first agent in minutes.