Open SWE is LangChain’s open-source framework for building internal asynchronous coding agents, modeled on the systems large engineering organizations run in production. Instead of an interactive pair programmer, Open SWE agents work like virtual teammates: you file a task, the agent works in an isolated sandbox, and it comes back with a pull request.
Built on LangGraph and the Deep Agents architecture, Open SWE is designed for parallelism and safety. Each task runs in a secure, isolated sandbox with its own environment, multiple agents can work separate issues at the same time, and every change arrives as a reviewable GitHub pull request. Agents can be triggered from GitHub issues, Slack, or Linear, and they can research code, plan changes, write and test code, and respond to reviewer feedback. The framework is model-flexible, working with leading models from Anthropic, OpenAI, and others. It has gathered more than 11,000 GitHub stars since its 2026 release and is under active development, with updates in October 2026. The code is MIT-licensed.
Open SWE is free and open source under the MIT license. You cover your own sandbox infrastructure and model API usage.
Engineering organizations that want to build and run their own internal fleet of coding agents rather than relying on a closed SaaS product.