Agents need somewhere to live.
Dock builds persistent cloud computers for AI agents: real Linux machines that keep their state, branch safely, and leave a record people can read.
Running code is easy. Real work needs a computer.
An agent doing real work builds up a world around the task: cloned repositories, installed packages, build caches, running services, logged-in browsers, previews, and files it made along the way.
Most sandboxes throw that world away when the job ends, so every team rebuilds it on every run. Dock keeps it. Each agent gets a machine it can come back to, copy, and hand to a person when needed.
What we believe
- State is part of the product
- Repositories, caches, browser profiles, and installed tools should outlive the task that created them.
- Safety lives outside the machine
- Isolation, credentials, network rules, approvals, and deletion must not depend on the agent behaving.
- Work should be easy to inspect
- You should see the commands, files, connections, snapshots, and cost behind a result without piecing logs together.
- Parallel work needs a family tree
- Branches should be cheap, independent, and always traceable to the snapshot they came from.
What Dock is not
Not a model. Not an agent framework. Not another throwaway code runner.
Dock is the computer underneath them. Bring your model, your agent, and your workflow. Dock gives the work an isolated place to continue, branch, and recover.
Your hardest workload is the useful one.
Tell us about it. We are onboarding teams in stages during the private beta.