Weston OS: an AI research agent for city government
Built inside a working municipal government, for the people who answer the public's questions all day. Presented with the City's permission; described in detail, not demoed.
Institutional knowledge trapped in PDFs, code books, and meeting videos
Every city hall runs on the same three questions: what will this permit cost, what does the building code actually say, and what did the commission decide. The answers existed, but they lived in fee schedules with stacked surcharges, code sections that reference other code sections, and years of commission meeting recordings nobody has time to scrub through.
Staff answered from memory and experience. That works until the experienced person is on vacation, retires, or is the one asking. As the IT leader responsible for the city's technology, I watched hours disappear into lookups that a machine should handle, and there was no product on the market built for this. So I built it.
An agentic research platform with government-grade answers
Weston OS is an agentic AI system: it doesn't just answer, it researches, keeps records of its work, and shows where every answer came from. In government, an answer without a citation is a liability, so citations were a hard requirement from day one.
Fully automated research across city clerk records, with agentic record keeping of what was searched and found.
Complete permit fee calculations with every surcharge itemized. Staff see the full breakdown, not a lump estimate.
Building code questions answered with proper citations to the governing sections.
Search across commission meeting records with video timestamps, so "when did we vote on that" ends with a link to the exact moment, not an afternoon of scrubbing footage.
Agentic workflows, not a chatbot wrapper
The platform runs on agentic AI workflows built with Claude and the Model Context Protocol: retrieval over municipal sources, tool use for fee mathematics, and structured record keeping of every research run. The design principle throughout is verifiable over impressive: a slightly slower answer with a citation beats an instant answer you can't defend at a public meeting.
It was also built with the constraints a real government imposes: municipal records rules, conservative security posture, and users who are experts in their departments, not in prompting. The interface asks for the question, not the magic words.
The product category existed in my head before it existed in the market
Companies now sell AI agents to local governments for exactly these workflows, and that category is growing fast. Weston OS is my proof that I understand it from both sides: I was the government buyer for 17 years, and I'm the builder who made a working system inside one, with real constraints, real records rules, and real staff using it.
If you're building or deploying AI for government, this is the experience I bring to your team on day one.
Weston OS is an internal City of Weston system. This case study is published with permission. No city data, credentials, or interfaces are reproduced here.