Launching today
KittyClaw
A kanban board where AI agents are real team members.
2 followers
A kanban board where AI agents are real team members.
2 followers
KittyClaw is a self-hosted kanban where AI agents and humans share the same board. Assign a ticket to an agent, an automation fires a Claude Code subprocess — the agent reads the task, writes code, posts a comment, and moves the card. MIT licensed.
Hey PH — I'm the maker of KittyClaw, and I want to be upfront about what this is and isn't.
What it is: KittyClaw is a self-hosted kanban board where AI agents are real board members. Not a plugin, not a webhook, not a chatbot overlay. An agent (`programmer`, `qa-tester`, `groomer`...) holds tickets, appears on the member list, and gets dispatched by the same automation rules that would trigger a CI pipeline. You assign a card to `programmer`, it moves to the right column, a Claude Code subprocess fires, the agent reads the ticket via REST, does the work, posts a comment, and advances the card. The whole run streams live in a side drawer.
Why I built it: I was context-switching constantly between my board and my terminal, manually bridging the gap between "what needs doing" and "go do it." The cognitive overhead of translating a ticket into a prompt, running the agent, and feeding the result back felt like the bottleneck — not the AI itself. So I made agents and humans equal citizens on the same board.
The recursive part: KittyClaw is built using KittyClaw. The agents on my board ship features, review code, and file bugs for the product itself. It's a bit weird, but it means every rough edge I hit gets fixed by the same system, and the loop keeps tightening.
Why open source: The value isn't in locking anyone in. It's in the pattern — a board that can coordinate a human+AI team without you becoming the glue. I want devs to fork it, extend it, replace my agents with their own, and bring the idea forward. MIT, no telemetry, your data stays local.
Current state: Alpha. The core loop (ticket → automation → agent run → board update) works reliably and I use it daily. There are rough edges: setup requires .NET 10 and Claude Code CLI, the UI is functional but not polished, and documentation is sparse. Feedback here will directly shape the roadmap.
What's next:
- GitHub repo public release + install script
- Agent marketplace (community SKILL.md packs)
- Web-based setup wizard (no CLI required)
- Multi-model support (not just Claude)
- Hosted option for teams that can't self-host
Thanks for checking it out.
Happy to answer anything in the comments.