kabana
Beta · launching soon

The kanban board that runs your roadmap while you sleep.

Drop a card. Your agent does the work. You review and approve when it's done.

Your AI account, your code Branch per card Watch your agent work in real time
Works with Codex recommended Claude Code Cursor CLI soon·Connects to GitHub
Why this exists

AI coding tools today are single-player.

You open a terminal, type a prompt, and watch. The work lives in chat windows that close at the end of the day. No shared backlog, no record of what shipped, no way to check in from anywhere but the chair you're sitting in.

No backlog. No plan.

Your tasks live in chat windows, sticky notes, and tabs you forgot to close. There's no view of what's in flight, what's done, or what comes next.

Tied to your keyboard.

You pay for an AI subscription around the clock, but it only works when you do. Every meeting, every break, every hour of sleep is paid time the agent isn't shipping code.

No record of the work.

Every prompt starts fresh. What you asked, why, what the agent decided. All of it lives in scrollback until the window closes.

What changes

Make AI coding multiplayer.

Move the agent out of your terminal and onto a board. The work gets a place to live, runs while you're away, and leaves a trail anyone can follow.

A real backlog.

Drop cards in To Do, watch them flow through In Progress to In Review. Your roadmap is a place you can look at, not a feeling you have.

Runs while you sleep.

Your CLI claims cards on your machine the moment you queue them. Walk away, go to bed, come back to a PR. Your AI subscription is finally earning its keep.

Every card is a record.

Brief, conversation, agent activity, and the branch it produced, all attached to the card. Find what you asked six weeks ago in two clicks.

How it works

From a card to a commit, in three steps

You stay in the board. Your agent does the typing. The work shows up as commits in your repo, ready to review.

1

Drop a card

Open the board, type what you need built, drop it into To Do. Pick the agent for the job, or let the board default decide.

2

Your agent picks it up

Kabana hands the card to the agent running on your machine. It knows your project from the brief and works on a fresh branch, never on your main code.

3

Watch it work, then review

You see what the agent is doing as it happens. When the work is done, the card moves to In Review and you decide what to merge.

Built around how you work

Your code. Your account. Your call.

Most AI coding tools want to take over: host your repo, proxy your model calls, mark up the bill. Kabana plans the work and shows the activity. Everything that touches your code happens on your machine, on the subscription you already pay for. You ship faster without handing the keys to anyone.

Your code stays on your laptop

The agent edits and commits in your local repo. We never see your source. Only the cards, comments, and the activity you choose to share.

Your AI subscription, your billing

Use the AI subscription you already pay for. Kabana doesn't proxy model calls and doesn't charge per token. Agent costs go to your AI provider, not to us.

Branch per card, no surprises

Every task gets its own branch off your base. Your main code stays untouched until you decide to merge.

Watch it work in real time

See every file the agent reads, every command it runs, every commit it makes. Straight from your machine to the board, as it happens.

Pick the right agent per card

Heavy refactor? Quick fix? Choose the agent that fits the job per card, or set a default for the whole board.

Reply on the card to continue

If the agent gets stuck or you want changes, leave a comment. The next run picks up the same branch with the full thread as context.

Early users

What the people shipping with it say.

Borrowed credibility beats self-promotion. The slots below get filled with real quotes from beta users as they land. Want yours here? Sign up and start shipping.

placeholder
Add a real beta-user quote here. One to three sentences works best.
Beta user
Role and company
placeholder
Pull out the line that captures the moment they realised it was working. Specific is better than generic.
Beta user
Role and company
placeholder
Keep one quote about the agent running while they sleep, one about reviewing later, one about parallelism.
Beta user
Role and company
Alan
Hey, I'm Alan.

I have my own business, I'm married and we have a 6 year old little girl.

My mornings are spent with my daughter, my afternoons working on my business, at night as an indie hacker.

I don't sleep too much, and I think it's affecting my memory. I also don't have much free time to work in my apps, my Claude tokens almost never hit its limits.

Building apps is what I do every day after my 9 to 5. It's not my full time job yet, but it's what I'm working toward. The hardest part isn't the code. It's the hours.

Side-project time is whatever's left after the day job, after dinner, after my daughter's asleep. By then I'm tired and the AI agent in my terminal is asking me clarifying questions instead of shipping.

I'm building kabana because I need it. I want to drop a card before I close my laptop and wake up to a PR. I want my Claude subscription doing something while I'm asleep. I want a backlog I can look at on my phone instead of a thread of prompts in a terminal that closed yesterday.

If you're shipping after 9 to 5 too, I'd love for you to try it.

Pricing

Pricing

Bring your own AI subscription. Kabana never proxies model calls, so you only pay for the kanban itself.

Individual
$3/mo
  • 3 kanban boards
  • 1 concurrent agentHow many cards can run at the same time. A slot can hold the same agent on different cards (say, two parallel Claude Code runs) or different agents on different cards (Claude on one, Codex on another). With one slot, cards run in priority order, the rest queue in To Do.
  • Claude Code agentBring your own Claude account. The agent runs on the subscription you already pay for, so any model usage goes to your Anthropic bill, not kabana.
  • Codex agentsoonBring your own OpenAI account. The agent runs on the subscription you already pay for, so any model usage goes to your OpenAI bill, not kabana.
  • Cursor agentsoonBring your own Cursor account. The agent runs on the subscription you already pay for, so any model usage goes to your Cursor bill, not kabana.
  • GitHub flow: branch per card + PRRequires setup. Connect your GitHub account in Settings and pick a repo for each board. The agent then commits to its own branch per card and pushes a pull request you can review in GitHub.
  • Approve & merge from kabanasoonComing with the kabana GitHub integration. For now, kabana opens the PR and you merge in GitHub yourself, then click 'Mark as done' to move the card to Done.
  • Kabana MCP
Get started

Cancel anytime.

Coming soon

More plans, more horsepower

Here's what's already on the roadmap.

More concurrent agents

Run cards in parallel across separate worktrees so two agents can ship two PRs at the same time, with no risk of stepping on each other's branches.

More boards per account

Higher tiers for founders, agencies, freelancers, and teams juggling multiple products or client repos from one workspace.

Auto-merge on green CI

Approved cards wait for GitHub Checks to pass, then merge themselves. Pairs with a kabana bot identity for repos that require approving reviews.

Review agent

An AI reviewer reads each card's diff the moment it lands in In Review, posts a verdict, and drafts the redo prompt if it spots an issue.

Team workspaces

Multiple people on one Kabana account, with audit history of who approved what and which agent shipped it.

Tell us what you need

Have a use case we haven't shipped yet? Drop a note in the changelog and it shapes what comes next.

FAQ

Common questions.

Does kabana see my code?

No. Your code stays on your laptop. The kabana CLI runs locally and edits your repo. The server only ever sees the cards you create, the comments you leave, and the activity stream the agent emits. No source files, no diffs.

Which AI agents work with it?

Claude Code today. Codex CLI and Cursor CLI are next. The kabana CLI auto-detects what you have installed on your machine and uses it to claim cards.

Do I need my own AI subscription?

Yes. Kabana doesn't proxy model calls and doesn't bill per token. Your agent runs on the subscription you already pay for (Claude Max, ChatGPT Plus, Cursor, etc.). Kabana's price covers the kanban only.

How much does it cost?

$3 per month, billed monthly. That includes 3 boards and 1 concurrent agent. There's no free tier and no per-seat pricing today.

Do I have to use GitHub?

No. Kabana ships three workflow modes: PR (opens a GitHub pull request when the agent finishes), direct (commits straight to the base branch), and nogit (skips git entirely). Pick per folder, change it later.

Can I cancel anytime?

Yes. Cancel from the Stripe billing portal in /settings. Your subscription stays active through the end of the period you already paid for, then ends. No auto-renew tricks.

Stop watching cards. Watch your repo move.

Sign up, install the CLI, drop a card. From zero to your first agent commit in under five minutes.