Running Konduktor in Docker would make things a lot harder as far as installation and management goes, and it would make Konduktor more limited.
That’s because we’d need to set up gh, Claude Code, and git inside the container, manage persistence and upgrades, the debugging story would be more difficult, and Konduktor agents wouldn’t be free to leverage the host system to achieve their goals. Konduktor agents running freely can install dependencies and explore the system in order to test things and unblock themselves.
Running directly on the host system means you can SSH in, and whatever plugins you install for Claude Code are immediately available to agents. Upgrading Claude Code is easy, and so is updating permissions for the gh CLI. A Konduktor agent might also install useful software that benefits other agents (if you let it).
