LangChain open-sourced Open SWE, a coding agent framework distilling architectural patterns from Stripe, Ramp, and Coinbase's internal AI dev tools.
LangChain released Open SWE, an MIT-licensed open-source framework for building internal coding agents, built on Deep Agents and LangGraph. The framework codifies architectural patterns observed across production deployments at Stripe (Minions), Ramp (Inspect), and Coinbase (Cloudbot). Core components include isolated cloud sandboxes, curated toolsets (~500 tools at Stripe), Slack-first invocation, rich context loading from Linear/GitHub/Slack, and subagent orchestration. A LangSmith Sandboxes waitlist is now open for the execution environment layer.
Open SWE gives engineering teams a production-validated architecture for internal coding agents without reverse-engineering what Stripe or Coinbase built. The LangGraph + Deep Agents stack handles subagent orchestration, sandbox isolation, and tool curation — the three hardest parts of agentic infra to get right from scratch. The MIT license means you can fork, strip, and rewire any component to fit your existing CI/CD or repo structure.
Clone the Open SWE repo this week and map its tool manifest against your team's most-repeated dev tasks (e.g., writing tests, triaging GitHub issues). If 3+ tasks match, wire up the Slack invocation layer to your existing workspace and run a controlled pilot on a non-critical repo.
Go to the Open SWE GitHub repo, open the tool configuration file, and paste the tool list into Claude.ai with this prompt: 'I have a [Node.js/Python/Go] codebase. Which of these tools are redundant for my stack, and what's missing for handling PR reviews automatically?' You'll get a prioritized tool audit in under 2 minutes.
Tags
Signals by role
Also today
Tools mentioned