# Brick and Mortar AI

> A self-service data platform for the Twin Cities: every public record a city already emits, joined into one warehouse and free to download -- parcels, lot lines, recorded sale prices, assessed values, reviews, Reddit, census profiles and business counts by trade. No account. CSV, XLSX, JSON or GeoJSON. A custom AI reads the same warehouse and is a free research beta at /model/. Built by a small team in Saint Paul, MN.

brickandmortar.dev turns a city's public records into a model an AI reasons over directly. The corpus behind it, as shipped: 25,203 real Google reviews across 75 restaurants, bars, cafes and taprooms (59 of them deep enough to rank against); 1,131,534 county parcels mapped across all 7 metro counties (Hennepin, Ramsey, Dakota, Anoka, Washington, Scott and Carver), with 1,632,823 shared lot lines and 182,350 road segments, so it can answer what a property is next to rather than only what it is; 868,994 recorded sale prices in Minnesota, which publishes them and most states do not; and federal employment, wage and consumer-price series for the metro running back to 1990, including every metro-wide employment contraction since then measured for depth and for how many months recovery took (3 of them so far).

The platform at brickandmortar.dev is a three-level funnel -- who you are, what you are looking for, and an export screen that hands you your slice as a file. 15 roles route to 61 datasets, 51 of them downloadable today; the ones that are not wired to an export yet show the real column list of the file they will be, say why, and take an upvote. There are also 40 tools (9 built), which invert the join -- you bring the private half, your own numbers, and the public record answers about them. Anyone can buy reviews and anyone can scrape a county; nobody else has them joined for one metro, which is why the join is the moat and the data itself is free.

Every downloadable dataset also has its own page under /datasets/, listing its real column names, its row count, the counties it actually covers and the cuts it can be filtered to. Coverage is not uniform and each page says so rather than averaging: it runs from one county to all 7.

The research beta at /model/ is a chat door onto the same warehouse, and is not the way in to the product -- the platform is. The visitor there is never asked who they are or where they are -- they pick a finding that interests them and the door routes on that. Asked about a restaurant it answers questions a general-purpose chatbot cannot: where a restaurant's rating actually sits in the real local distribution, what genuinely drives negative reviews ranked by frequency, which shifts carry the most risk, and how the owner replies that work differ from boilerplate. Asked about a property it serves the person who rents or is about to rent a space: what the building last sold for, what the county says it is worth and how far that runs under real sale prices, how long the owner has held it, what shares its lot line, and whether the local economy behind that address is actually growing.

## Pages

- [Home](https://brickandmortar.dev/): The self-service data platform. Pick a dataset, filter it to you, download the file -- 1,131,534 parcels across 7 metro counties, 1,632,823 shared lot lines, 868,994 recorded sale prices and 25,203 business reviews behind it. Free, no account, no rate limit.
- [Research model](https://brickandmortar.dev/model/): The chat door onto the same warehouse -- one question, then the records that fit the answer. Paste any address in the 7 metro counties. A free research beta, not the core product.
- [All datasets](https://brickandmortar.dev/datasets/): Every downloadable dataset in one table -- rows, columns and the counties each one actually covers.
- [Export API](https://brickandmortar.dev/api/export): The downloads without the page. `?dataset=&scope=&columns=&format=` returns CSV, XLSX, JSON or GeoJSON; call it with no arguments for the catalogue. Every file carries the source and the layer's own stated limits.
- [System Card](https://brickandmortar.dev/system-card/): What the model is built from, layer by layer, with each layer's own record count, source and stated limits -- plus what the public record cannot answer at all. The page to read before believing a figure it gives you.
- [Licence](https://brickandmortar.dev/license/): CC BY 4.0 over every dataset here -- free to use commercially, the one condition is crediting Brick & Mortar with a link. Also names the datasets we hold back and why.
- [Case Studies](https://brickandmortar.dev/case-studies/): Real results for real local businesses -- rate intelligence, financial clarity, and AI enablement.
- [About](https://brickandmortar.dev/contact/): Who's behind Brick & Mortar -- a small team in Saint Paul, MN.

## Licence

Every dataset listed below is published under Creative Commons Attribution 4.0 International (CC BY 4.0) -- https://creativecommons.org/licenses/by/4.0/. You may redistribute and adapt them, including commercially. The only condition is attribution: Brick & Mortar — https://brickandmortar.dev/

Two datasets are deliberately NOT published as files, on rights rather than coverage grounds: `occupants`, `search-visibility`. Both are derived from Google, and we cannot grant a commercial redistribution right we do not hold. They are analysed and their findings are public; the rows do not ship. Minnesota Secretary of State bulk business filings are held back on the same principle.

## Datasets

Each has its own page with the full column list, the counties it covers and its download links in CSV, XLSX, JSON and (where the rows locate to an address) GeoJSON.

- [What shares a lot line with a property](https://brickandmortar.dev/datasets/adjacency/): 115,652 rows, 28 columns.
- [Who is listed at every commercial address](https://brickandmortar.dev/datasets/businesses/): 88,433 rows, 20 columns.
- [Walk a corridor, address by address](https://brickandmortar.dev/datasets/corridors/): 44,151 rows, 53 columns.
- [What sold, and for how much per square foot](https://brickandmortar.dev/datasets/sales/): 34,302 rows, 24 columns.
- [Healthcare providers with a federal NPI, statewide](https://brickandmortar.dev/datasets/providers/): 32,257 rows, 18 columns.
- [Which properties FEMA puts in the floodplain](https://brickandmortar.dev/datasets/flood/): 28,354 rows, 14 columns.
- [Who the landlord is, on a licensed rental property](https://brickandmortar.dev/datasets/landlords/): 23,303 rows, 35 columns.
- [What the 911 calls around here are about](https://brickandmortar.dev/datasets/emergency-calls/): 22,123 rows, 20 columns.
- [How a neighbourhood differs from the typical one](https://brickandmortar.dev/datasets/census-tracts/): 10,797 rows, 12 columns.
- [Who owns it, and what else do they hold](https://brickandmortar.dev/datasets/owners/): 9,251 rows, 13 columns.
- [What a bank actually lent, loan by loan](https://brickandmortar.dev/datasets/sba-loans/): 8,822 rows, 32 columns.
- [Cleanup and storage-tank files, by address](https://brickandmortar.dev/datasets/contamination/): 7,149 rows, 18 columns.
- [Who does the work here, and how to write to them](https://brickandmortar.dev/datasets/contractors/): 6,588 rows, 14 columns.
- [Who holds a licence to trade here](https://brickandmortar.dev/datasets/licences/): 6,553 rows, 11 columns.
- [Subsidised homes, and when the subsidy runs out](https://brickandmortar.dev/datasets/subsidised-housing/): 1,500 rows, 23 columns.
- [Business counts by trade, five vintages](https://brickandmortar.dev/datasets/trades/): 318 rows, 13 columns.
- [Who lives here, and what they pay](https://brickandmortar.dev/datasets/county-profile/): 24 rows, 6 columns.
- [Building stock by era](https://brickandmortar.dev/datasets/building-stock/): 12 rows, 7 columns.

## Entity

- Legal name: Brick and Mortar AI LLC (Minnesota). Trading name: Brick & Mortar AI.
- Service-area business based in Saint Paul, Minnesota. No walk-in location; serves the United States.
- Contact: aidan@brickandmortar.dev

## Notes

- The review corpus is Upper Midwest and skews full-service. Complaint patterns, owner-reply behavior, and day-of-week effects generalize to other markets; absolute rating percentiles are regional and are presented as panel figures, not national ones.
- Review excerpts used by the model are anonymized and are never attributed to a named business.
- Coverage stops at the metro's edge -- the 7 Minnesota counties named above, and nothing beyond them. Inside that boundary the layers are not all the same width: parcels, lot lines, roads and recorded sale prices cover all 7, while the census profiles and everything built on reviews cover Hennepin and Ramsey only. Where a layer is narrower it says which counties it has rather than answering from the ones it knows best.
- Greater Wichita commercial real estate was live here and is not now. The Sedgwick County corpus still exists; there is no door to it on this site, so nothing here should be read as covering Kansas.
- The employment, wage and price series are federal (BLS) and metro-wide. They describe the Minneapolis-St. Paul metro, never a neighbourhood, and are labelled that way in every answer.
- This file is generated by scripts/build_llms_txt.py from the shipped substrates. Do not hand-edit it.