GitHub – luccast/crabwalk: 🦀 Crabwalk 🦀 Real-time companion monitor for Clawdbot agents.

If you’ve ever run AI agents and felt that quiet unease of not really knowing what they’re doing behind the scenes, you’re not alone. Watching logs scroll by isn’t the same as actually understanding behavior. That’s where **Crabwalk** comes in, a small but thoughtful open source project that feels like it was built by someone who’s been there too.

Crabwalk is a real time companion monitor for Clawdbot agents, created by luccast. Instead of guessing how an agent moves from prompt to tool call to response, you can literally watch it happen. Live. Visually. Almost like standing behind someone’s shoulder while they work, without being annoying about it.

The core idea is simple and very human. You open a browser, point it at your local setup, and see your agents operating across WhatsApp, Telegram, Discord, or Slack in a live node graph. Thinking states appear. Tool calls branch out. Response chains connect. It’s the kind of clarity you wish you had the first time an agent went off script at 2 a.m. and you had no idea why.

What makes this especially interesting is how approachable it feels. Crabwalk runs locally, works alongside the Clawdbot gateway, and can be spun up with Docker Compose if that’s your comfort zone. Open your browser at http://localhost:3000/monitor and you’re in. No ceremony. No heavy setup rituals.

Under the hood, it uses familiar modern tools like ReactFlow, TanStack Start, tRPC, and Framer Motion. But you don’t need to care about any of that to appreciate the result. The interface is there to help you think, not to impress you.

Looking ahead, tools like Crabwalk hint at a future where monitoring AI systems feels less like debugging and more like observation. Calm. Visual. Understandable. If you’re curious to explore it yourself, the full project lives here and is worth a slow look: https://github.com/luccast/crabwalk