PolitiClaw: building a local-first political co-pilot with OpenClaw 🦞

OpenClaw is an open-source agent framework. Its plugin system is how you get from "chatbot" to something that runs in the background, remembers what you've told it, and acts on a schedule. I spent some time building a plugin called PolitiClaw on top of it, and most of what I know about OpenClaw I learned by doing that.

I'll start with a quick tour of OpenClaw for folks who haven't seen it: gateway, tools, skills, cron, memory. Then the plugin SDK: What definePluginEntry actually does. How to register tools the agent can call. How to ship skills as Markdown files a user can override without touching your code. And the parts the docs don't cover, which I had to work out by reading the SDK source.

Then PolitiClaw: It watches federal legislation, scores reps against stances a user has declared, and drafts outreach letters that can be pasted into an email client. It runs entirely on a user's machine, no telemetry, no cloud piece. I'll walk through the bits that were interesting to build and the design call I keep coming back to: the whole thing is an LLM application, but a lot of the real work was deciding where not to let the model near the data. I'll explain why.

No OpenClaw experience needed. A little TypeScript helps but isn't required.

• All levels
• Development
• AI/ML
• Community