Does this sound familiar?

An applicant is at the counter with a question about an irregular lot.

You know the answer is in the ordinance. You also know that getting to it means pulling the zoning ordinance for the district table, checking the definitions section for how lot width is measured, cross-referencing the subdivision ordinance for the depth-to-width rule, and confirming whether sanitary sewer is available. Twenty minutes, minimum. Probably forty.

So you give them the version you're confident in, tell them you'll confirm, and add it to the pile.

Multiply that by every walk-in, every phone call, every pre-application meeting, every plan review. Then consider that the veteran reviewer who holds most of this in their head is three years from retirement, and the new hire is asking you the same question for the fourth time this month.

That's the problem this solves.

What is the Tool this guide builds

The Municipal AI Agent is a configuration method — a way of setting up a general-purpose AI assistant so that it behaves like an experienced municipal code official reasoning only from your jurisdiction's own adopted documents.

You load your ordinances and codes. The assistant treats them as the entire universe of authority. Every answer comes back tied to a named document and a specific section, table, or page.

Ask it whether a use is permitted in a district and it works the same sequence a good reviewer works: property and jurisdiction → district and overlays → existing and proposed use → permitted, special exception, or prohibited → dimensional, parking, access, landscaping, and site-plan standards → subdivision, utility, drainage, building code, accessibility, and fire triggers → which permits, boards, inspections, and fees apply → what facts are still missing.

Then it returns the answer in a fixed format built for the permit record:

Issue → Documents Consulted → Findings → Determination (Compliant / Noncompliant / Indeterminate) → Open Items & Needed Facts → Recommended Next Steps

That output can be attached to a permit file, sent to an applicant, forwarded to the City Attorney, or read into the record at a public meeting.

THE PART THAT MATTERS MOST

The most valuable thing this system does is refuse to answer.

You already know what happens when you ask a general-purpose AI a code question. It answers. Confidently. Every time.

It cites an IRC section from a code year you never adopted. It describes a setback standard from a model ordinance instead of yours. Asked something your ordinance is genuinely silent on, it manufactures a plausible answer rather than admitting the gap — because plausibility is what it was built to produce.

For a code official, that isn't a productivity tool. That's a liability generator with a text box.

The Municipal AI Agent is configured around three non-negotiable rules that invert this:

1. Cite precisely. Every claim names the document and the exact locator. No claim without a citation. If it can't cite it, it doesn't say it.

2. "Indeterminate" is a first-class answer. When the documents are silent, when the scope of a standard genuinely doesn't reach the situation, or when essential facts are missing — the correct output is Indeterminate, accompanied by a specific list of what would be needed to decide. Not a guess dressed as a finding.

3. Surface conflicts, don't resolve them. When two adopted documents conflict, it identifies both and explains the impact — without inventing a hierarchy your documents don't actually establish.

There's also a specific technique in the blueprint that makes rule #2 work in practice: pairing a broad "what do the documents say about this" search with a narrow "which section actually governs this situation" search. That two-query pattern is what lets the system correctly return Indeterminate on ambiguous scope questions instead of over-answering them. It's the single most-copied element of the setup, and it's fully documented.

This is the difference between a novelty chatbot and a tool your City Attorney is comfortable with.

WHAT IT HANDLES

Zoning & Land Use

Use permissibility by district, special exception and variance pathways, dimensional and density standards, lot buildability, accessory uses, parking and surfacing, signs, nonconformities, site-plan requirements, administrative procedure.

Building Code & Plan Review

Residential and commercial review, decks (guards, handrails, footings, ledgers), foundation dampproofing vs. waterproofing and drainage, sidewalk construction specifications, commercial reroofing scope, demolition, and generic permit conditions placing construction responsibility on the contractor.

Permits, Fees, Registrations

building, driveway, sidewalk, demolition, location-improvement, electrical, plumbing, and mechanical fees; registration requirements; owner-performed vs. registered vs. specialty contractor distinctions.

Correspondence and records

Plan-review comment letters, applicant and developer explanations, design-guidance documents, contractor correction notices, internal memoranda, Council-facing narratives, and drafted permit conditions — in factual, neutral, records-appropriate language.

WHAT IT LOOKS LIKE IN PRACTICE

The vague inquiry that became a defensible determination

A resident asked whether an irregular polygon lot in a single-family district was buildable. Staff asked the system to identify every question that would need answering to decide.

It produced a 22-question intake checklist — organized by property identification, lot status and history, dimensions, frontage, use, utilities, and environmental constraints — each item tied to the ordinance provision that made it necessary.

When the applicant's answers came back (polygon lot, ~11,533 sq ft, 97-foot frontage), the system evaluated them against the district tables and returned Compliant — buildable, while flagging three verification items: sanitary sewer confirmation, survey-verified width at the setback line, and lot-of-record status against the depth-to-width rule.

One tool turned an unstructured citizen question into a complete intake instrument and a determination that could stand in the file.

The question it correctly refused to answer

A resident wanted to store an RV in the side yard. The system confirmed it was permitted with conditions — side or rear yard, five-foot setback, resident-owned, with limits on temporary front-driveway parking. Cited.

Then the follow-up: can the pad be gravel?

The system returned Indeterminate. The surfacing standard's scope may not reach single-family yard storage, and the accessory-use section is silent on surface material. It recommended obtaining a written interpretation from the enforcement official.

That is the correct answer. A general-purpose AI would have picked one and sounded certain.

The analysis that didn't get run on the wrong track

An inquiry came in about opening a coffee shop on a vacant parcel. Instead of producing a use analysis, the system led with the controlling gap: the parcel's zoning district has to be established first — because a restaurant-type use is permitted by right in some districts, requires a special exception in others, and is prohibited in some.

It then laid out the provisional end-to-end approval sequence, contingent on that answer. Staff didn't spend an afternoon analyzing the wrong district.

The memo that got read at a Council meeting

An applicant in an alley-vacation matter claimed he'd been told verbally that he was "good to go." The system drafted a formal memorandum to the Common Council establishing that verbal conversations do not constitute formal authorization, attached the permit report and review timeline, and laid out the factual record without editorializing.

It was shortened on request, finalized, and read into the record.

WHAT IS INCLUDED

The bundle is everything required to stand up your own instance. Downloadable, editable, and yours permanently.

1. The Advisor Instruction Set — the complete persona and directive block, copy-paste ready. The expert-official role, the non-negotiables, tone calibration, and the rules that keep it from behaving like a chatbot.

2. The Knowledge Base Assembly Guide — exactly which documents to load, how to name and structure them, how to handle amendments and supersession, and the file-organization decisions that determine whether citations come back accurate or vague.

3. The Standardized Determination Format — the Issue → Documents Consulted → Findings → Determination → Open Items → Next Steps template, plus the secondary formats for direct answers, checklists, review comments, and drafted correspondence.

4. The Evaluation Sequence — the step-by-step reasoning order the advisor works for any project-specific question, written out so it can be dropped straight into your configuration.

5. The Accuracy Guardrails — the cite-precisely, Indeterminate, and conflict-surfacing rules, with the exact language that makes them stick.

6. The Two-Query Search Method — the broad/narrow pairing that produces correct Indeterminate results on scope-ambiguous questions.

7. The Worked Example Library — real prompts and real outputs across zoning, plan review, permits, and correspondence. This is what you hand a new hire so they can see what good looks like before they write their first query.

8. Platform Setup Guides — Claude and ChatGPT. The same architecture, built out for both. You are not locked to a vendor, and you can run it on whatever your staff already has.

9. The Video Walkthrough — the full build, start to finish, so you're following along rather than interpreting.

10. Limitations and Disclaimer Language — the advisory-scope and human-review language, pre-written, for the conversation with your City Attorney. Most municipal AI projects die in that meeting. This one is built to survive it.

PRICING - $500. One time.

No subscription. No per-seat cost. No annual renewal. No vendor contract, no data processing agreement to negotiate, no implementation consultant.

And no procurement process. In most municipalities, $500 sits below the department-head discretionary threshold — which means you can buy this on a card today without a purchase order, a council agenda item, an RFP, or a six-month approval cycle.

That is deliberate. The reason good tools don't reach small planning departments has very little to do with whether the tools work.