Devboxes
AI coding agents that work on your repositories, in containers you control.
Devboxes runs AI coding agents against your Git repositories. You describe a
task and dispatch it — from the dashboard, the devboxes
CLI, or a Slack mention; an agent works in an
isolated container while you watch the transcript live, answer its questions,
and review its diff — and it pushes the finished work back to your repository,
typically as a pull request.
Your first run takes four things: a repository connection, a project built from it, a runner that can serve your model, and the dispatch itself. The quickstart walks that whole path.
The shape of it
- Projects pair one repository with the container image its runs execute in.
- Blueprints define the steps an agent works through — yours, or the built-in ones.
- Agent sessions are where you watch a run, steer it, and review its changes.
- Runners execute runs: Devboxes Cloud, or your own machines through the Devboxes CLI.
How these fit together — and where your API keys live along the way — is the overview.
Where to start
User guide
Everything you do in Devboxes: connect repositories, create projects, dispatch and steer runs, manage credentials — and run tasks on your own machine with the Devboxes CLI.
Working as a member
Invited to an existing organization? What you see, who dispatches runs, and how to follow the work live.
Feature catalog
For the team building Devboxes: each feature's behavior contract, the tests that prove it, and the decisions behind its shape.