The public record, read at the source.
Relm Data is a library of 2,500+ municipal, county and state open-data sources across 11 Northeast states — permits, certificates of occupancy, fire, violations, inspections, complaints and zoning. Ask about an address and Relm queries every source that covers it, keeps the raw records, and answers from those records only, with a citation on each one.
Key Capabilities
Where the library reaches today
Every figure here is generated from the source catalog and the latest live sweep, dated 2026-10-12. Counts are sources, not records — one source can hold millions of rows.
- New York1,427
- Maryland735
- Pennsylvania352
- Massachusetts333
- Connecticut172
- New Jersey107
- Delaware77
- Vermont71
- Rhode Island66
- New Hampshire40
- Maine30
What an ask reads, and what else is on the shelf
Seven topics are asked by default — the ones that decide whether a building is what the listing says it is. The rest of the catalog is browsable and queryable one source at a time.
Asked by default
- Building permits323 sources
- Certificates of occupancy6 sources
- Fire incidents & inspections146 sources
- Code violations115 sources
- Inspections131 sources
- 311 & complaints72 sources
- Zoning & land use175 sources
Also in the library
- Assessments328
- Housing & rent registration215
- Transportation189
- Crime & public safety188
- Licenses129
- Property tax96
- Energy & benchmarking87
- Flood zones84
- Health80
- Parcels80
- Environmental78
- Utilities52
- Evictions15
- Other municipal datasets821
Read from the portal that publishes it
A handful of the sources behind the numbers. Each one is queried on the publisher's own open-data portal, and every record Relm returns links back to it.
- DOB NOW: Build – Approved Permitsdata.cityofnewyork.us · NY · Building permitssource_id f16b07e62fe12d3a
- DOB Certificate Of Occupancydata.cityofnewyork.us · NY · Certificates of occupancysource_id 8d3814f0c2d7d3a0
- DOB Violationsdata.cityofnewyork.us · NY · Code violationssource_id ea539011dc5e4493
- 311 Service Requests from 2020 to Presentdata.cityofnewyork.us · NY · 311 & complaintssource_id b7d6df1a53bf5de7
- Approved Building PermitsInspectional Services Department · MA · Building permitssource_id 76670149bdf39c38
- Building Permits: New Constructiondata.cambridgema.gov · MA · Building permitssource_id 1da999add3f54e53
- Building Permit Data System (BPDS)Baltimore · MD · Building permitssource_id 8615750da3d180ca
- Commercial Permitsdata.montgomerycountymd.gov · MD · Building permitssource_id b73f8a4cc391aff7
- Building Permits 20200101 to CurrentCity of Hartford · CT · Building permitssource_id 2eabe5dd03e11c67
- Code Enforcement 20200101 to CurrentCity of Hartford · CT · Code violationssource_id af74b41871c76bc0
- Department of Inspections and Standards Permits 2009-2018data.providenceri.gov · RI · Inspectionssource_id 87104a82f09ddd4b
- Philadelphia L&I permits (Carto SQL)Philadelphia · PA · Building permitssource_id 7edca5aa465e2c0a
In the app, by API, or by an agent
The same library behind every Intel Report is exposed as eight REST endpoints and eight MCP tools. Both are read-scope operations. Browsing the library is free; pulling records costs 2 credits per call (per source answered in a batch) and an ask 20.
// Any MCP client connected to https://api.relm.ai/mcp/
resolve_public_record_sources({
"address": "350 5th Ave, New York, NY 10118",
"topics": "permits,violations"
})
// → sources covering the address, ranked city > county > state
ask_public_records({
"question": "Any open permits or violations?",
"address": "350 5th Ave, New York, NY 10118"
})
// → answer + raw records + one citation per record + coverage# Which sources cover the address (nothing fetched)
curl "https://api.relm.ai/v1/data/library/resolve?address=350+5th+Ave,+New+York,+NY+10118" \
-H "Authorization: Bearer relm_sk_..."
# Raw records from ONE source, with citations
curl "https://api.relm.ai/v1/data/library/f16b07e62fe12d3a?address=350+5th+Ave,+New+York,+NY+10118&limit=25" \
-H "Authorization: Bearer relm_sk_..."The eight operations: browse the library, read a source’s catalog entry, resolve which sources cover an address, pull raw records from one source or up to ten at once, ask a question across every matching source, ask about a whole city or county, and search records already fetched. Full reference in the API docs and the MCP server guide.
How an ask works.
Give Relm an address and a question. Four things happen, in order:
- Resolve: the address is geocoded and matched to the sources whose publisher covers it — city first, then county, then state — for the topics you asked about
- Fetch: each source is queried directly on its own portal, by point-and-radius where the dataset has a location, by street address where it does not; personal fields are dropped before anything is stored
- Cache: the rows land in a shared record cache with a per-topic freshness window, so the next ask about the same address is answered from records already fetched
- Answer: the data agent writes its answer from the fetched records only and may cite nothing that is not in them; the raw records, a hash over them and a coverage report travel with every answer
A source that could not be reached is reported as exactly that — never as "nothing found".
Inside every Intel Report.
Relm Data is the public-records step of an Intel Report. When the agents brief a property, this step asks the library about open permits, certificates of occupancy, code violations, fire incidents, inspections, complaints and zoning for the address, and the findings land in the report with the same citations you would get calling the library yourself. Teams can add their own notes and corrections on top of the records; the raw rows stay untouched underneath, with version history.
Built for agents as much as for people.
Every library capability is available three ways. In the app, through an Intel Report. Over the REST API, as the /v1/data/library endpoints. And over the Relm MCP server, as five tools an AI agent can call directly: browse the library, resolve which sources cover an address, pull raw records from one source, ask a question across all of them, or search records already fetched. The REST and MCP surfaces share one implementation and one set of docs, so what an agent sees is exactly what the API returns.
Honest about coverage.
The library is swept against sample addresses in every state, and the sweep is what keeps it honest: a source whose endpoint has gone dark is marked unreachable and skipped, a portal that changed shape fails a scheduled check rather than silently returning nothing, and the coverage figures on this page are generated from the catalog and the latest sweep rather than typed by hand. Today the library covers the Northeast; coverage grows one publisher at a time, and each one is verified live before it is added.
Related Features
See Relm Data in Action
Start your first property analysis and experience how Relm Pro transforms underwriting.