Deploy your agent as a function, get a computer for it.
1. Write an agent as a TypeScript function. Deploy it. We run the loop, the sessions, streaming etc
2. Every session runs on a real Linux machine.
3. Your tools and MCP servers run on that machine - clone a repo, run ffmpeg, drive a browser, install anything.
4. Sessions are durable: can be steered mid-run, hibernate when idle, resume where they left off.
5. No model keys in your runtime. Bring your own or use our managed gateway.
the durable session part is what sold me - hibernate and resume means I'm not paying for idle compute while an agent is just waiting on something, and not losing installed packages either. a real Linux machine per session with ffmpeg/chromium/git preinstalled beats the usual sandboxed-function approach where you're constantly reinstalling deps on every cold start.
What needs improvement
pricing isn't on the launch page yet, so I can't tell what a long-lived hibernating session actually costs versus a short-lived one - would help to see that upfront rather than having to sign up to find out.
vs Alternatives
looked at plain Fly.io machines and Modal first, but both push you toward stateless containers or app-shaped sandboxes. OpenComputer's session-as-a-first-class-resource model fits an agent that runs for hours, not seconds, better than either.