KittyClaw

The problem

When you're a solo dev driving one or more AI agents on your projects, you need a place to track progress: who's doing what, what's blocked, what's ready for review. A kanban, basically.

Except existing tools all have issues:

  • Jira, Azure DevOps are built for 50-person teams. Heavy setup, complex permissions, painful APIs to interface with an agent.
  • GitHub Projects is better, but the API goes through GraphQL with a not-always-intuitive data model, and you need to juggle tokens for every agent.
  • Linear, Height are SaaS — you depend on a cloud, you pay per user (including your agents), and you bend your workflow to theirs.

None are built for solo + agents, and all demand more energy to configure than to actually use.

The solution

KittyClaw is a local kanban that does more than track tickets: it orchestrates your Claude Code agents. Create a ticket, an automation engine spots when it's ready, launches the right agent as a claude subprocess, streams its output into a live drawer next to the ticket, and commits its work when it's done. No cloud, no subscription, no setup — run.bat (Windows) or ./run.sh (macOS/Linux) and you've got a board on localhost:5230.

What makes it different:

  • A built-in AutomationEngine — triggers (ticketInColumn, gitCommit, boardIdle, subTicketStatus…), conditions (labels, priority, ticketAge…) and actions (runAgent, moveTicketStatus, commitAgentMemory…), declarative in automations.json per project. The harness that replaces your homemade dispatcher.
  • Named agents as first-class citizens — programmer, groomer, producer, qa-tester, committer, code-janitor, evaluator, each with their own mentions, memory (memory.md versioned in the repo) and history.
  • A complete REST API — everything the UI does, an agent can do. Auto-generated OpenAPI docs at /api/docs, always in sync with the running server.
  • Smart onboarding — on first launch, KittyClaw detects whether claude and git are on your PATH. If anything's missing, it tells you up front instead of letting you wonder why nothing fires.
  • Zero friction — one SQLite DB per project under %APPDATA%/KittyClaw/, no auth, no network config. Up and running in 30 seconds.
  • Yours to shape — open source and modifiable. Every week I add what I need. The story of these updates captures how the tool sculpts itself through use.

What it changes, concretely

When you work with AI agents, you want them to be able to:

  • See what needs to be done, without navigating a complex interface
  • Act — create a ticket, move it, comment, mention someone
  • Communicate — with you, with each other, with no ambiguity about who's speaking
  • Run by themselves — without you having to launch them by hand every time

KittyClaw does all of this natively. On Aekan, 13 specialized agents work in parallel on a single board, orchestrated by the AutomationEngine. You create a ticket assigned to @programmer, thirty seconds later the agent is on it — code written, commits made, ticket back in review. When an agent finishes, a committer can take over to commit and push. When a ticket hits Done, an evaluator wakes up to check quality. The anatomy of a ticket that runs itself walks through the full pipeline.

What about Claude Code?

Claude Code is excellent for coding with an agent. But when you're running a real project, it hits a wall: you jump from one conversation to another, you lose the big picture, you forget who's working on what, you repeat yourself three times because nothing is tracked beyond the session.

KittyClaw fills exactly that gap. You step into supervisor mode: you see all tasks in flight, you prioritize, you unblock agents, you review when it's ready. Agents themselves run as claude subprocesses with sessions persisted across runs (--resume), find their context on the ticket via the API, and write their memory to .agents/{agent}/memory.md versioned in the repo.

The result: you move 10x faster, because everything is tracked, organized, and persistent. No more repetition, no more decisions lost in chat, no more agents going in circles. You direct, agents execute.

Who it's for

  • You work solo (or as a pair) and want project tracking without ceremony.
  • You drive one or more AI agents and want to fit them into your workflow.
  • You want to own your tool — not depend on a SaaS that changes its roadmap.

Prerequisites

Without claude and git you can launch the app, but agent runs and auto-commits will fail. KittyClaw tells you on first launch.

Get started

git clone https://github.com/Ekioo/KittyClaw.git
cd KittyClaw
run.bat        # Windows
./run.sh       # macOS / Linux

Runs on http://localhost:5230 with hot reload. Markdown API docs at /api/docs.

Learn more