LangChain Open SWE: Open-Source Framework for Building Internal Coding Agents
LangChain's open-swe is trending on GitHub with 454 stars per day, reaching 5,900+ total stars. It's an open-source framework for building asynchronous coding agents modeled after enterprise implementations at companies like Stripe, Ramp, and Coinbase.
Open SWE lets organizations deploy AI coding bots that can be invoked through Slack, Linear, or GitHub. Agents operate in isolated cloud sandboxes (Modal, Daytona, Runloop supported) where they access repositories, execute commands, run tests, and automatically create pull requests. The system uses a curated set of ~15 focused tools rather than accumulating hundreds, and supports spawning subagents for parallel work.
Built on top of LangChain's Deep Agents framework using LangGraph, it provides an upgradeable foundation rather than a monolithic codebase. Repositories can include an AGENTS.md file to encode conventions, and task context flows naturally from Linear issues or Slack thread history.
The project is MIT-licensed with 20+ active contributors.
https://github.com/langchain-ai/open-swe
← Back to all articles
Open SWE lets organizations deploy AI coding bots that can be invoked through Slack, Linear, or GitHub. Agents operate in isolated cloud sandboxes (Modal, Daytona, Runloop supported) where they access repositories, execute commands, run tests, and automatically create pull requests. The system uses a curated set of ~15 focused tools rather than accumulating hundreds, and supports spawning subagents for parallel work.
Built on top of LangChain's Deep Agents framework using LangGraph, it provides an upgradeable foundation rather than a monolithic codebase. Repositories can include an AGENTS.md file to encode conventions, and task context flows naturally from Linear issues or Slack thread history.
The project is MIT-licensed with 20+ active contributors.
https://github.com/langchain-ai/open-swe