Open Agents is a platform for spawning autonomous coding agents that live in the cloud, work against your repositories, and ship — while you are elsewhere, entirely.
The agent does not live inside your machine. It lives outside of it — in a sandbox with a filesystem, a shell, a git, and a clock of its own. You describe the work. It clones the repo, opens a branch, writes the patch, runs the tests, and when it is satisfied it commits, pushes, and files the pull request under its own byline. You close the laptop. The agent keeps working. When you come back, the work is waiting for review.
Works autonomously.
A ToolLoop with file ops, search, bash, task delegation, and first-class subagents. Choose the model. Set the effort. It plans, it patches, it verifies its own work.
Works anywhere but yours.
Isolated Vercel microVMs with git installed, ports exposed, and dev servers that actually run. Each session gets its own. Hibernates on idle, resumes from snapshot.
Works past the browser.
Durable Workflow SDK runs survive restarts, reconnects, and requests. Close the tab; the agent does not. Come back; the stream resumes. No work is ever lost mid-step.
Fork the repo, deploy your own copy, and spawn your first agent in the time it would take you to read the rest of this page.