Let agents run freely without putting your machine at risk.
Switch a project to Docker or Podman in one click, and its commands run in a container that sees only your project. KodaCode picks the base image, so there’s nothing to set up manually.
Inside the container, the agent sees your project and the folders you choose to share. The rest of your machine stays on the other side.
Open a terminal inside the agent’s container to run tests and debug exactly where it works, or a shell on your own machine from the same menu.
Full speed for your agents. Peace of mind for you.
Each project can run in Docker, Podman or straight on your machine, or just follow your default. KodaCode picks a base image by looking at the project (Rust, Go, Python or Node), or uses a Dockerfile you keep in the project's .koda folder, or one you choose. Try out an unfamiliar repo in a container and keep the one you know well on your machine.
| The project | Mounted at /workspace, where the agent reads, builds, tests and edits |
| A build cache | Kept per project, so downloads and builds aren't repeated from scratch each time |
| Folders you allow | Only the ones you've added in Settings, and nothing else from your machine |
| Skills | Shared read-only, so their scripts can run but never be rewritten |
| Everything else | Not there: your home folder, your other repositories, the folder next door |
When a project runs in a container, the terminal opens inside it, so you get the same tools, files and limits as the agent. Click + for another container shell, or for a shell on your own machine in the project folder. Each tab is labelled DOCKER, PODMAN or HOST, and the panel can sit under the session or float over it.
| In a container | Changes apply directly and commands aren't stopped by default, because the box holds only the project and what you share |
| On your machine | Changes ask first, and so does anything that reaches outside the project |
| Either way | Each is a switch you can flip for the project · how approvals work |