background-agents

Background-agents lets you fire off AI coding agents that go work on their own. You hand one a task, it spins up a sandboxed dev environment, clones your repo, does the work, and opens a pull request while you do something else. It's the self-hosted, open source version of what Cursor's background agents and tools like Devin sell as a subscription. MIT licensed.

Free to use is not the same as free to run. This is a multi-service system: a Next.js app, Cloudflare Workers, Python pieces, a GitHub App you install, plus an account with a sandbox provider like Modal, Daytona, or Vercel Sandbox to actually execute the agents. You also bring your own model API keys, Anthropic or OpenAI. So the software costs nothing, but you're paying the sandbox provider and the model bills, and standing up real infrastructure to do it.

It's early and maintained mostly by one person, so treat it as capable-but-young, not battle-tested. Solo devs and small teams who want to experiment with autonomous coding without a per-seat Devin or Cursor bill will get the most out of it. Teams that need reliability and support today are better off paying for the hosted product for now.

The catch is the ops burden versus the payoff. Standing this up is real work, and background agents are still a young idea that produces mixed results. You're trading a subscription for infrastructure you maintain and API bills you watch. Worth it if you value control and want to learn the pattern. Not worth it if you just want PRs to show up reliably.