Relm Pro is opinionated: every number in the system should be either traceable to an external source or explicitly marked as an assumption. This page describes the kinds of sources we draw from and how we cite them — without naming individual vendors, since coverage and providers evolve.
Public records
- County assessor and treasurer records — assessed value, tax history, land/improvement split, owner of record.
- Federal demographic surveys — population, income, age cohorts, education at block-group resolution.
- Federal labor statistics — local employment, wage growth, sector breakdown.
- Federal rent benchmarks — used as a sanity check on AI-generated rent assumptions.
Listing and rental platforms
Aggregated listing data feeds comps, asking rents, days-on-market, and unit photos. We use both broker-listed and ILS (internet listing service) data, depending on what's available for the asset class.
The units pipeline has a v1 (Stable) and v2 (Beta) path — see Units v1 vs v2. The v2 path uses an agentic crawler that can build a unit list from scratch when listing data is weak.
Institutional property data
Tax history, ownership-of-record, and historical sales come from a normalized property graph that aggregates assessor and recorder records across most U.S. counties. Coverage is strong in tier-1 metros and most Sun Belt markets and thinner in some rural counties; when the graph is thin, Relm transparently falls back to the underlying county record and surfaces a different citation.
Maps and points of interest
Distance, walkability, and POI counts (transit, schools, retail, employers, recreation) come from a major maps and places API. Citations link to the underlying place records.
AI / language models
We use multiple frontier model providers for different tasks — reasoning-heavy steps go to one, structured extraction to another, embedding to a third. We do not train any model on your data. See AI in Relm.
Your uploads
The single highest-quality source for a deal you're underwriting is the documents you upload. When a rent roll is present, the rent roll wins for unit count, unit mix, occupancy, and in-place rents — every other source is used only as a sanity check. The same precedence holds for any property fact derivable from a rent roll.
Upload formats we handle:
- PDF (scanned or text)
- XLSX / XLS / CSV (rent rolls, T-12s, P&Ls)
- DOCX / TXT (offering memorandums)
- Image formats (JPG, PNG) for site plans and unit photos
How we cite sources
Every value in a section card has a citation chip — click it to see the underlying source URL or document page. If a value couldn't be sourced, it's marked as Assumed with the rationale Relm used. This is non-negotiable: a number with no citation and no rationale should never appear in the UI. If you spot one, report it as a bug.
What we do not provide
- Title work — we don't pull title or commitments. That's still your title company.
- Structural / mechanical inspection data — out of scope.
- Tenant credit / rent payment history — except where present in your rent roll.
- Loan quotes — we model debt at assumptions you control, not at quoted rates.
If a deal needs a third-party perspective on top of all this, the Relm Report layers a human analyst on top of the same data.