Open SWE: LangChain's Open-Source Internal Coding Agent

Open SWE is LangChain's open-source answer to the internal coding agents that power Stripe's Minions, Ramp's Inspect, and Coinbase's Cloudbot. It's an asynchronous coding agent framework: you submit a job from Slack, Linear, or GitHub, and Open SWE runs it in an isolated cloud sandbox and opens a pull request with the result — no babysitting required.

Why internal coding agents?

Elite engineering orgs don't just use coding assistants — they embed agents directly into their workflow: Slackbots, CLIs, and web apps that engineers trigger without leaving their tools. Two key insights from the teams that built them: tool curation matters more than tool quantity, and every task runs in its own isolated environment so the blast radius of a mistake stays small. Open SWE is built on exactly these principles, composed on Deep Agents and LangGraph.

Architecture at a glance

Key features

• Trigger jobs with a mention — @openswe in Slack, Linear, or GitHub
• Instant acknowledgement with 👀 and in-thread status updates
• Message the agent mid-run; it picks up follow-ups at its next step
• Parallel tasks, each in its own isolated cloud sandbox
• Automatic draft PRs linked back to the original ticket
• Subagents for parallel subtasks, plus a web dashboard and experimental desktop app