open-swe

Why we included this project

Open SWE suits teams that have outgrown chat-in-the-IDE helpers and want a coding agent able to work on its own for longer stretches. It follows the architecture Stripe, Ramp, and Coinbase built for their internal agents: each task runs in an isolated cloud sandbox with shell access, the agent works through a small set of tools for shell, web requests, and Git, and it can open pull requests and post progress to Slack or Linear without a human watching every step. Because it is built on LangGraph and Deep Agents, subagent orchestration and middleware hooks are already there. The agent also honors AGENTS.md conventions your team already keeps, so existing rules carry over without extra wiring. Plan for some setup time: this is a framework to adapt, not a finished product, so you will wire your own repos, sandbox provider, and workflow hooks.