LoanCite extracts every figure, cites its source, and runs the rules — so when a note buyer or regulator asks why, you hand them one record instead of excavating inboxes.
When an investor samples your loan, they don't ask about the process — they ask about the number on page 7. "We ran it through AI" is not a defense.
Bank-statement, P&L, 1099, asset-depletion — each method involves steps a reviewer can dispute. If the math isn't shown, the number isn't defensible.
An investor updates a matrix on a Tuesday. You have 40 files in flight. You find out when a loan gets kicked back, not before.
Most AI tools send the full document to an external model. That's borrower PII leaving your environment, and a per-page bill you can't predict.
Bank-statement, P&L, 1099, asset-depletion — each has its own income method, its own edge cases, its own audit exposure. LoanCite runs the math the way an underwriter does and shows every step.
Every figure is cited to its source page. Every rule has a verdict. Every exception is logged and resolved by a human. The full decision chain is hash-chained and exportable on demand.
Rules are config — your toggles, severities, and thresholds, not ours. Recognized documents parse without an AI key. Everything stays inside your deployment. A real engineering team handles the setup.
Generic extraction tools get data out of documents. LoanCite gets data out and proves where it came from.
| Capability | LoanCite | Per-page | Agents | Platforms |
|---|---|---|---|---|
| Built for non-QM instruments (bank-stmt, P&L, 1099, asset-depletion) | ● | ○ | ◐ | ○ |
| Income computed with cited, reproducible math | ● | ○ | ◐ | ○ |
| Citation on every figure, traced to source line | ● | ◐ | ◐ | ○ |
| Human in the loop on every exception | ● | — | ○ | ◐ |
| Hash-chained, tamper-evident audit trail | ● | ○ | ◐ | ◐ |
| Decision-ready binder, figure-level & guideline-version pinned | ● | ○ | ◐ | ○ |
| Key-less extraction, no per-page meter | ● | ○ | ◐ | ◐ |
| GSE LL-2026-04 vendor-governance pack | ●Roadmap | ○ | ○ | ○ |
Compared to common categories of AI lending tools, not specific vendors.
Start with document processing or jump straight to underwriting. Add the other when you're ready.
Each agent connects to your existing LOS, cloud storage, and email. They run independently or together — start where the pain is biggest.
Fannie Mae's new lender letter requires an AI governance framework — and it explicitly covers the vendors behind your tools, not just your own use. LoanCite ships the governance pack, audit trail, and human-in-the-loop layer out of the box, with an exportable record you can hand to an investor or regulator on the spot.
Informational, not legal advice. LoanCite does not claim that human-in-the-loop or any specific tool is legally mandated.
Every rule has an owner, a version, and a change history. Config-only — no code changes required.
Blocking exceptions route to your underwriter. Nothing auto-clears. The approval is logged to the record.
A single PDF + JSON export covering your AI use, your vendor (us), and every decision on a given file.
LoanCite runs inside your infrastructure. Borrower documents never touch our servers unless you explicitly configure that.
Recognized document types parse deterministically — no external model call, no per-page cost, no PII leaving your environment.
Each lender gets their own extraction config, rule set, and branding. Isolation is architectural, not just access-controlled.
Built to SOC 2 standards. We'll tell you exactly what we have and what's in progress — not what looks good in a deck.
The live demo runs an actual bank-statement loan through extraction, income math, and rule checking — in real time. Then book time to see it against your own guidelines and document types.
No — and that's intentional. LoanCite sits alongside Encompass, ICE EPC, and most other systems. The file we produce lands in your LOS; your team works the same way they do today, just with a complete, cited package already in front of them.
Never. LoanCite prepares the file and reports on the rules. The underwriter decides. That's not a limitation — it's the design. Human sign-off is what makes the decision defensible.
With private deployment, it stays in your infrastructure — full stop. The key-less parser handles standard documents without calling any external model. The AI only sees what can't be parsed deterministically, which is typically scanned or non-standard pages.
Yes. Rules are config-only — your thresholds, your severities, your toggles. For custom income methods or document types we haven't seen before, our engineering team handles the build. We've done this for shops with unusual P&L structures and non-standard investor matrices.
It likely does if you use any AI or ML in your origination process — and it covers your vendors, not just your own tools. LoanCite ships the governance framework, audit trail, and human-in-the-loop layer as part of the standard product. We can't give legal advice, but we can show you exactly what we provide and let your counsel decide.
Most pilots are running in two to four weeks. Standard income methods and common document types are configured out of the box — custom work takes longer, but we scope that before you commit.