Multi-agent AI workflow board powered by KaibanJS and BitNet.
Describe any task in plain English. AI Kanban automatically:
- Plans a team of specialized agents suited to the goal (2–6 agents)
- Runs each agent sequentially with real back-and-forth reasoning — each agent asks a clarifying question before producing its output
- Passes context between agents so each one builds on the previous agent's work
- Visualizes everything in real time on a Kanban board (Todo → Doing → Done)
- Frontend — single HTML file, no build step, no dependencies to install
- KaibanJS — provides the Agent/Task/Team orchestration and ReAct reasoning loop
- BitNet — 1-bit LLM running on CPU, handles all inference (planning + agent reasoning)
- Cloudflare Worker — proxies requests to BitNet with CORS headers and adapts responses to OpenAI format so KaibanJS can consume them
| Layer | Technology |
|---|---|
| Agent framework | KaibanJS |
| LLM | BitNet (CPU inference) |
| Proxy | Cloudflare Workers |
| Frontend | Vanilla HTML/JS |
Open index.html in a browser (or serve from any static host). No API keys required — BitNet is free and open.
The Cloudflare Worker source is in worker/index.js. It handles:
- CORS for browser requests
- SSE → JSON translation (BitNet streams SSE, KaibanJS expects OpenAI JSON)
- Skeleton extraction — strips KaibanJS's ReAct prompt scaffolding and replaces with targeted calls so BitNet produces clean structured output