What it does
A purpose-built engine for loan package review.
Loan Intelligence turns complete or partial loan packages into review-ready outputs. Each capability below is a step in the same deterministic flow.
Capability list
Core engine
Process loan packages in minutes, not hours.
Loan Intelligence ingests 10–30 source documents as a single package object, extracts borrower, property, income, and collateral facts, and returns review-ready outputs. Tested packages of 1–30 documents can process in under three minutes.
Policy intelligence
Bring your own credit policy.
Upload your credit policy and Loan Intelligence extracts your LTV, DSCR, FICO, eligibility, exclusion, and document rules. Every package is evaluated against the active policy for a clear inside_box, conditional, or outside_box status.
Gap control
See what is missing before the underwriter does.
Every package returns a source-linked checklist of required and expected documents. When a new document arrives, the package can be reprocessed without rebuilding the methodology.
Verified calculations
DSCR, LTV, and payment calculations you can inspect.
When required inputs are present, Loan Intelligence calculates deterministic metrics in code. Results are tied to the source pages that produced them.
Connected intake
Let loan packages arrive from Drive or Gmail.
Connect a controlled Google Drive intake folder or receive documents by email. Loan Intelligence converts incoming files into source documents without treating generated outputs as new inputs.
Agent-native
Give your AI agent a loan package engine.
Connect Claude, Cursor, Codex, or any MCP client through a single JSON-RPC endpoint. Agents can quote, submit, and retrieve results with the same methodology as the workspace.
FAQ
Capability questions that matter in production.
These answers focus on package behavior: incomplete files, credit policy, deterministic calculations, source-linked outputs, and where Loan Intelligence fits beside existing systems.
Yes. Incomplete packages are expected in real lending workflows. Loan Intelligence extracts what is present, calculates only metrics with enough inputs, and returns missing or unclear items as review tasks. It should not invent absent facts just to make the package look complete.
A workspace can store an active credit policy. Numeric rules such as DSCR, LTV, FICO, page-count, and document requirements are locked into deterministic checks where possible. Eligibility, exclusions, and remediation are evaluated against extracted facts and citations. The output is inside_box, conditional, or outside_box decision support for a reviewer to inspect.
When the required inputs are present, calculations run in code with source-linked input values. If an input is missing or ambiguous, the output should show the gap instead of filling it with an unsupported assumption. That is the difference between a review engine and a plausible narrative.
The public contract returns package status, session state, result manifests, source references, extracted facts, warnings, tasks, and generated output links. The response is intentionally product-shaped: enough for humans and agents to act, without exposing queue names, raw staged content, cache heads, or internal billing idempotency keys.
The package can be reprocessed with the new source set. The quote reflects the current source document and page counts, and the result should show updated facts, missing items, and policy status. Connected intake paths are designed to separate original sources from generated outputs so reports do not become new billable source documents.
No. Loan Intelligence is the intelligence layer around the package. It can sit beside a loan origination system, Drive folder, email intake flow, or agent workflow and return source-linked outputs those systems can use. The lender still owns recordkeeping, approvals, compliance policy, and final credit decisions.
Ready to process your first package?
Start with credits today. Upload files, call the API, or connect through MCP.
Get started