Source: github.com/diggerhq/opencomputer
Team members mention @Computer in a Slack thread with a task, and the agent works on its own computer. Teammates can watch and steer it while it runs, and the finished work arrives as a pull request. The whole stack can run in a user's own cloud account or on OpenComputer Cloud.
The gateway, called Calvin, runs on a single VM and starts sessions with Claude Code, Codex or OpenCode, depending on configuration. Each session is isolated with rootless Podman, and its state is written to local disk before being persisted to a git-backed store. The remaining components run on Cloudflare Workers and Durable Objects, and support for Fly is planned.
In the project's benchmark, all four PRs opened with CI green in 56 seconds, against 185 seconds for Devin. First push came in 24 seconds versus a 62-second average for four popular cloud agents. The hosted version offers a three-day free trial with 600 work minutes, and users can bring their own ChatGPT or Codex subscription.
Why it matters if you build agents
Teams that want a coding agent working from Slack can self-host the full stack, including the VM and Cloudflare layer, instead of depending on a hosted service. The speed figures come from the project's own benchmark, so they should be tested against your own repositories before you rely on them. Self-hosting also means you maintain the locked-down outbound network setup and the Podman sandboxes yourself.
Get stories like this every weekday
One short email with the day's agent tooling news.