Daikonbun is a task board that runs itself. Drop in a rough task; a pipeline of AI agents refines, plans, reviews, builds, and verifies it — moving the card across the board and pausing for you only at the gates that matter.
Be first through the gate. One email when it ships — no noise.
A card moves left to right through typed stages. Each stage is worked by a specialized agent role; each review gate produces a verdict and a score. Approve to advance, or send it back.
Turns a rough note into a clear task with acceptance criteria.
›Writes the implementation plan and decision log.
›Scores the plan. You confirm — or it loops back.
›Implements the change and writes the tests to hold it.
›Reviews the diff for correctness and quality.
›Runs the full suite. Green ships; red loops back.
›Merged, with the whole story on the card.
Every plan, review, and test result is signed back onto the card. A resumed session — or a teammate — can reconstruct exactly what happened.
It pauses at plan and code review. Full autonomy is opt-in, per run. You decide how long the leash is.
Agent roles are defined by function, not vendor. A routing table maps each role to whatever LLM you choose.
One docker compose up. Your data, your keys, your machine. Credentials encrypted at rest, never returned in plaintext.
The pipeline refines before it builds and reviews before it ships. A circuit breaker stops a card that can't satisfy its gates.
Columns, cards, drag-to-move, a live run feed. Supervising many parallel agent runs should feel like mission control.
If you already run coding agents and want a durable, reviewable process around them — this is for you.
One email when Daikonbun ships. Early self-host access, the docs, and the day-one release.