KeepCouncil

AI reads the record. We show our work.

KeepCouncil uses AI to read city documents. It writes short, plain answers. We don’t hide that. Every answer shows where it came from and how sure we are.

From city documents to your answer

  1. Collect

    We gather what the city posts. That means agendas, minutes, and meeting packets.

  2. Read

    We pull out the words and page numbers. We mark any page we could not read well.

  3. Organize

    We sort what we find into meetings, decisions, votes, and people. Code counts the votes. AI does not guess them.

  4. Answer

    When you ask, we look up the facts first. Then AI writes a short answer from only those facts.

Two kinds of AI help. OpenAI writes the answers. Jev AI, from typesafe.ai, sorts the pages, rates how well each one was read, and checks every answer.

What the confidence marks mean

You’ll see these next to answers, votes, and summaries.

  • Very confident

    The record says it plainly, and every source agrees.

    Example: A roll call vote written out in the minutes.

  • Confident

    The record backs it up. We put it in plain words, or joined facts from more than one place.

    Example: A short summary of a whole meeting.

  • Low confidence

    The record is missing parts, hard to read, or unclear. Check the source before you rely on it.

    Example: A vote where the minutes don’t list who voted.

We check, then check again

  • Every claim links to the page it came from.
  • Votes and dates come from the organized record before any AI writes.
  • A second AI checks each sentence against its sources before you see it. Sentences that don’t match are removed.
  • If that check can’t run, the answer says “Not fact-checked” and gets a low mark. Read its sources before you rely on it.
  • Pages we couldn’t read well get a low mark. We never pass them off as clean.

What the AI won’t do

It won’t pick candidates or tell you how to vote. It won’t guess why people did things, and it won’t take sides. It only retells what the public record says.

AI summaries

Some pages have a short summary written by AI. It always says “AI summary” and has a confidence mark. We check each sentence against the city’s documents and remove any that don’t match. The official title is always right next to it.

Topics

Each city has its own short list of topics, like Stormwater or Parks. A computer check puts each agenda item under one topic, and sometimes one more. It picks only from the list and never makes up a new topic. The small bars on a topic show how sure it is. Meeting steps like roll call are not counted.

Community notes

If a vote, a summary, a topic, or a page looks wrong, open its … menu and add a community note. You need an account, so each note has a name on it. Most notes show up right away. Some wait until we read them, such as notes from new accounts. Then we check each one, and its badge shows where it stands:

  • Waiting for review Only you can see it for now. We read it before anyone else can.
  • Not yet reviewed Just posted. We haven’t checked it yet.
  • Verified The note is right. A fix is on the way.
  • Corrected The note was right, and KeepCouncil now shows the fix.
  • Not supported The record supports what KeepCouncil shows.

Every note keeps its history. A note never changes the city’s own documents.

City badges

Some notes and replies carry a city’s name in a badge, like this: Creve Coeur workspace member

It means the writer is in that city’s KeepCouncil workspace. To join, a person asks, and the contact on the city’s own website says yes. We don’t check anyone’s job. So the badge shows the city let them in. Each writer speaks for themselves.

The public record always wins

Our answers are a guide, not the official record. Every answer links back to the city’s own documents, so you can check for yourself.