Langroid is a Python framework to build multi-agent LLM-powered applications. It manages Agents with optional LLM, vector store, and tools to collaboratively solve problems, using a multi-agent paradigm.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Langroid is a lightweight Python framework to easily build LLM-powered applications by setting up Agents, equipping them with optional components (LLM, vector-store and tools/functions), assigning them tasks, and having them collaboratively solve problems by exchanging messages. It supports multi-agent collaboration and does not require using Langchain or other LLM frameworks.
How it works
Langroid provides Agent abstractions (e.g., ChatAgent) and Task constructs to orchestrate interactions between agents, with optional components such as LLMs, vector databases, and tools. The Quick Start example shows configuring an LLM via OpenAIGPTConfig, creating an OpenAIGPT model, and using Agents within Tasks to run multi-agent interactions. It also mentions a MCP Tools adapter to connect agents to MCP servers and convert server tools to ToolMessage instances.
Getting started
The README includes a Quick Start section with code snippets illustrating:
- Setting up LLM configuration:
llm_cfg = lm.OpenAIGPTConfig( ... ) - Instantiating an LLM:
mdl = lm.OpenAIGPT(llm_cfg) - Creating an Agent:
agent_cfg = lr.ChatAgentConfig(llm=llm_cfg) - Creating a Task and running it:
task = lr.Task(agent, name="Bot", system_message="You are a helpful assistant")andtask.run("Hello") - Example of a 2-Agent chat loop with teacher and student agents
Recent releases
Recent releases include updates up to 0.65.16 (2026-07-29) with notes such as: "0.65.16 — sloppy LLM JSON keeps all its fields again". Prior entries mention changes like rotating API keys for Vertex AI, SQL-based classification improvements, and dependency adjustments (e.g., unpinning LanceDB and PyArrow). See the release notes listing for details.
Traction
Stars: 4089 (as per FACTS). Forks: 390. Open issues: 76. Language: Python. License: MIT.