Relm Data is a catalogued library of government open-data sources — building permits, certificates of occupancy, code violations, fire incidents, inspections, 311 complaints, zoning and more — that Relm reads directly from the portal each city, county or state publishes. It is what fills the Public Records section of an Intel Report, and the same library is open to your own tools over the Relm API and MCP.
It is not a copy of someone else's property database. Every record you see was fetched from the publishing government's own open-data endpoint, is stored exactly as published (with applicant and owner personal details removed), and carries a citation back to the dataset it came from.
What it covers
Relm Data currently covers the Northeast: Connecticut, Delaware, Maine, Maryland, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island and Vermont. Across those eleven states the library holds more than 3,400 catalogued sources from roughly 400 government publishers, of which 2,400+ can be queried by street address or location — the ones a lookup actually reads. The live figures, a coverage map and a list of example sources are on the Relm Data feature page.
Coverage follows what governments publish, so it is uneven on purpose. New York City alone accounts for a large share; a small town that posts its permits as a PDF is not in the library. When nothing is published for an address, the section says so rather than guessing — see Honest about gaps below.
Seven topics are read by default
A lookup asks the topics that decide whether a building is what the listing says it is:
- Building permits — open and closed work, with issue dates and descriptions.
- Certificates of occupancy — the legal use and unit count on file.
- Code violations — open and resolved housing, building and property-maintenance violations.
- Fire incidents & inspections — fire department runs and inspection results where published.
- Inspections — building, elevator, boiler and similar inspection records.
- 311 & complaints — service requests and complaints logged against the address.
- Zoning & land use — zoning districts, overlays and land-use designations.
The rest of the library — assessments, parcels, licenses, evictions, flood zones, energy benchmarking, transportation, utilities and other municipal datasets — is browsable and can be queried one source at a time through the API or MCP.
How a lookup works
When an Intel Report runs (or when you ask the library a question by API), Relm does four things:
- Resolve. The address is geocoded and matched to the sources whose jurisdiction covers it — city first, then county, then state — for each requested topic. Nothing is fetched yet, so this step is cheap and is also how the API tells you what coverage exists before you spend a longer call.
- Fetch. Each matched source is queried on the publisher's own portal, by location (within about 60 metres) or by street address depending on what the dataset supports. Sources are queried in parallel, with per-portal limits so Relm is a polite consumer of public infrastructure.
- Cache. Records are stored as fetched, with personal fields removed and a hash over the record set so you can tell whether anything changed between runs. Repeat lookups inside a topic's cache window return quickly; a cold lookup on a dense city address can take up to a minute.
- Answer. For an Intel Report, the public-records agent writes a short summary from the records only. It can cite only
[n]indexes that exist in the record set — it cannot introduce a fact that is not in a fetched row.
The Public Records section in the workspace
On a property, the Public Records card shows how many records were found and from how many of the matched sources ("38 records from 9 of 14 official sources"), with a per-topic count. Open it for the full view:
- Records grouped by topic, newest first, each expandable to the full row as published.
- Coverage — which sources were tried, which returned records, which were empty, and any topic where nothing was found.
- A sources list — one citation per source, linking to the dataset on the publisher's portal.
- The summary written from the records, with its reasoning and the matching assumptions it relied on.
Correcting your copy
Public records are sometimes wrong or attached to the wrong parcel. You can edit a field on a record or remove a record from your organisation's copy — the record is marked Edited, and the change never alters the record as the government published it. If records are removed after the summary was written, the section flags that the summary's citation numbers may be out of date; re-run the section to rewrite it.
The section is optional in a run: if a portal is down or the address has no coverage, the rest of the Intel Report completes normally and the card explains what happened.
Honest about gaps
An empty result is a finding, not a failure. The coverage panel distinguishes three situations:
- No source covers the address — the publishing government does not expose this topic as data. The card says "nothing found for …" and names the topic.
- A source was reached and had nothing near the address — recorded as empty, with the source still cited so you can check it yourself.
- A source could not be reached — recorded as an error with the portal's reason. Relm re-sweeps the catalogue and retires endpoints that have gone away, so a persistent error usually means the publisher moved or removed the dataset.
A few sources are published behind paid portals or bulk licences. Those appear as locked with a record count and the note Available with Relm Data; their records are not fetched until the Relm Data add-on is enabled for your organisation.
A few others are public but shown only to a signed-in user, with no data endpoint at all — FDNY's FIRES portal behind an NYC.ID account, for example. Those appear as needs a portal login until your organisation connects an account under Settings → Third-Party Services; Relm then signs in, in an isolated browser, to read them. See Portal accounts.
What it costs
- In an Intel Report, the Public Records section is charged like other sections, by what it delivers: roughly two credits per record kept, plus a handful for the address match and the summary. Citations, coverage and the record hash are provenance and are never charged. An address with no published records costs only those few credits. See Credits.
- Over the API and MCP, browsing sources, reading a source's catalog entry and resolving an address are free. Pulling one source and searching fetched records cost 2 credits per call, and a batch read of several sources 2 credits per source that answered; asking a question, about an address or a whole city or county, costs 20 credits per call. Normal API rate limits apply.