We're the people who show up to the zoning meeting and read the model card. A community for applying AI to local democracy — carefully, transparently, and in public.
Founded by people who have personally transcribed a three-hour council meeting by hand. Never again, ideally.
A real Tuesday, roughly.
Most of the decisions that actually change your block — zoning, budgets, policing contracts, bus routes — get made in rooms with three residents in the audience and a clerk taking minutes nobody will read. That's not a people problem. It's a bandwidth problem, and it's exactly the kind of problem language models are good at, if you point them carefully and supervise the output.
So we built a place for the people already trying: civic technologists, city clerks, open-data nerds, and the one council member who insists on reading the whole packet.
Turning agendas, minutes, and hours of recorded testimony into plain-language digests — with every claim linked back to the timestamp it came from, so "the AI said so" is never the end of the sentence.
Making public-records requests, council archives, and permit filings searchable in the language residents actually use — not the language the filing cabinet uses.
Summarizing thousands of public comments by theme without flattening the minority report into a footnote. Clustering is a starting point for a human reader, never a replacement for one.
Turning capital plans and CAFRs into charts and explainers a resident can read on their phone before the meeting starts, not after the vote is taken.
Helping cities that adopt AI tools write the boring, essential paperwork first: procurement standards, audit rights, bias testing, and a named person who can turn the thing off.
Chapters meet monthly, pick one real problem in front of their own local government, ship something small, and write it up so the next chapter doesn't start from a blank page. No keynote speakers required — bring a laptop and a question.
A single ordinance, backlog, or budget line. Small enough to finish, real enough to matter to someone outside the room.
Prototype with actual public documents, actual civic staff feedback, and a running list of everything the model got wrong.
Publish what worked, what didn't, and the prompts and caveats — so the next chapter inherits a head start, not a rumor.
Robert's Rules exist so a room full of disagreeing people can still get business done. Ours exist so a room full of language models can too.
A model can draft the summary, the redline, the digest. It does not get a vote, and it does not get the last word.
Any tool that ranks, flags, or scores a constituent needs a named human who can override it and explain why.
Public decisions don't get made with private logic. At minimum, publish what the tool does and where it can fail.
A renewal date and a named owner, from day one — not a pilot that quietly becomes permanent infrastructure.
A clustering algorithm doesn't get to average dissent away. The smallest cluster still gets read into the record.
We'll take a dull, well-documented tool over an impressive demo that nobody can explain to a city attorney.
Whether you write policy, write code, or just write down what happened at the last meeting so someone else doesn't have to — there's an agenda item with your name on it.
Prefer to write first? The chair reads everything, eventually — meetings run long.
The mailing list is stored, not sold — no algorithm deciding who gets read first. Just a chair, and an inbox.