Kanbun

kanbun

A live race-board for autonomous coding agents.

Kanban boards weren't built for agentic development. Once you have more than three or four agents running at the same time — especially across different projects — the static "todo / doing / done" model falls apart. Cards pile up under "doing" and you can't tell which agent is actually moving, which one is stuck waiting on CI, which one crashed three steps ago, or how long any of them have been there.

kanbun replaces the kanban board with a **race-board**. Each agent gets its own horizontal lane. The phases of your workflow (plan → implement → PR → review → merge) become columns. Cards race left-to-right through the phases in real time, with a sub-progress bar per phase so you can see whether an agent is 5% or 95% through "implement". Cancelled, failed, and merged runs settle into their final column with the right colour.

Click any card → a drawer slides open with the linked Jira card on one side and a live tail of the agent's Claude Code transcript on the other.

Built with [Bun](https://bun.sh) + React + [Vaul](https://vaul.emilkowal.ski). ~1k lines, no build step, no framework.

**Pairs with [Archon](https://github.com/coleam00/Archon).** kanbun is just the UI — you need a harness like Archon to actually run the agents and populate the workflow database that kanbun reads from. Any harness that writes to the [schema below](#expected-sqlite-schema) will work, but Archon is what we built it against.


What it does

• Polls a SQLite workflow DB every 1.5s and renders one bar per run.
• Reconciles Archon-style run status against per-run final-status.json sidecars (see Sidecar schema) so a workflow that crashed after the PR merged still shows as completed.
• Click a bar → drawer opens with the linked Jira card (left) and a live tail of the agent's Claude Code session transcript (right), streamed via SSE.

Why "kanbun"

Kanban + Bun