Cloud Parts

Cloud agents

Agents that run each task on a machine in someone else’s cloud, take work from a chat, ticket or web app, and come back with a pull request.

How they work

A cloud agent gets its own computer per task: a VM or container with your repository cloned, its dependencies installed and, often, a browser. You start it from a web app, Slack, an issue tracker or the CLI, and it keeps working after you close your laptop. Devin, Codex cloud and Claude Code’s cloud sessions all work this way; OpenAI’s first Codex cloud release said tasks typically took 1 to 30 minutes and ran with internet access disabled.

The separate machine is both the point and the cost. It can’t damage your laptop, and you can run many tasks in parallel, but its environment has to be set up to match yours (setup scripts, secrets, network allowlists), and your code runs on the vendor’s infrastructure. Cognition, which makes Devin, frames the split this way: a local agent makes you faster, a cloud agent does the work while you’re not there. Several products now let you start a session locally and hand it to the cloud.

Technologies

What all ai coding agents have in common →

Sources

Improve this page on GitHub