GitHub integration and one-click Docker environments

Most demos of AI coding start with an empty folder. Most real work starts with a repository that already has four years of history, a Docker Compose file nobody fully understands, and a README that lies. This release is about the second kind.

GitHub, connected properly

Connect your GitHub account to an agent with OAuth. Browse your repositories and branches from the dashboard and clone the one you want directly into the agent's workspace. No personal access tokens to paste into a terminal, no SSH keys to copy around. The connection lives with the agent, and when you clone the agent later, its git identity and access come along.

Docker, detected and started

Once a repository lands, the agent looks for Docker Compose files and profiles. If it finds them, you get a button. Press it and the stack comes up on the agent's machine, with build output streaming into the dashboard and the exposed ports listed in the status bar so you can open the running app in the agent's browser.

• Profile selection when a project has more than one way to run
• Port mappings persisted per agent, so they survive restarts
• Bring the stack down just as easily when the job is done

Why it matters

The value of an agent is bounded by the environment you can give it. An agent that can clone your actual project, start your actual database, and hit your actual API is an agent that can fix the bug your customer reported this morning. That is the job.

From here on, when we say an agent can build something, we mean inside the codebase you already have.