Kanbots turns every Kanban card into a local coding-agent workspace

Kanbots is an open-source desktop Kanban app where each card can run its own coding agent. The project’s notable design choice is local-first storage: the SQLite database, configuration, and worktrees live in a .kanbots/ directory next to the repository, without a cloud account or hosted coordination server.

The idea lands directly in the current tension around coding agents. Parallel agents promise more throughput, but they also create more diffs, more branches, and more review burden. HN commenters quickly moved from “can this run many agents?” to “how much unsupervised work can a developer realistically review?”

There is also a useful tension in the Kanban framing. Traditional Kanban makes work visible partly so teams avoid taking on too much at once. A board that encourages one agent per card pushes in the opposite direction unless it also gives the human operator strong controls for work-in-progress, isolation, and review.

The local-first approach drew favorable attention because developer tooling often crosses trust boundaries. Keeping state beside the repo makes the system inspectable and portable. At the same time, commenters raised practical questions: each worktree may need its own services, ports, preview URL, database state, and visual test path.

That makes Kanbots interesting as a workflow experiment, not merely as another wrapper around model calls. Agent boards are becoming execution surfaces. The hard part is not launching more agents; it is turning their parallel work back into reviewable, mergeable, locally reproducible changes.