🔀 I built an AI agent router that picks the right model for the job — and it changed how I code 🦞

I got tired of one model doing everything badly.

💸 Expensive reasoning on simple searches.
🏃 Fast models hallucinating on architecture.
🚫 Rate limits with no fallback.

So I built a task router. One prompt, zero config hell.

⚡ What it does:

🧠 Smart routing — says "plan this", "debug that", "review this PR" and sends it to the specialist that handles it best
⚡ Instant response — you never wait at a blank screen. Agents work async, results appear when done
📊 Live status — every 2 minutes, a heartbeat tree shows exactly which agent is running, blocked, or done
🔄 Auto fallback — rate limit hit? Switches to the next best model. You never notice
🛡️ No hallucination — hard constraints: never invent files, never guess imports, always search first
💰Cost smart — simple tasks get cheap fast models, complex ones get deep reasoning. No waste

👀 What a session looks like:

🔀 Router

├── 🏗️ plan (Microservices arch) → [RUNNING]

├── ⚡ fast (Find auth module) → [FINISHED: src/auth/]

└── 📋 Summary: Architecture in progress, auth found

🛠️ Setup:

Drop 2 files in your OpenClaw workspace. That's it.