Mortgage document intelligence for the agentic era.
One API that turns raw loan documents into indexed, structured data. Split a 1,000-page loan package in seconds and classify a pay stub in a fraction of a second.
Built for the documents in a real loan file
Pay stubs
W-2s
1040s & K-1s
Bank statements
Appraisals
Closing Disclosures
Servicing mail
Lending and finance teams building on Addy
Read and understand mortgage documents in seconds.
Indexing is the first step in every loan pipeline, so it can’t be the slow one. Use advanced document intelligence models to deliver a fast and seamless lending experience.
1,000-page loan package
~9.8s
11 document types found
p.1 · p.500 · p.1000
Closing package 312
Bank statements 148
Tax returns 96
Disclosures 84
Title 72
Appraisal 54
Credit report 38
URLA / 1003 26
Pay stubs 18
W-2s 12
Other 140
// POST /split
{ "pages": 1000, "documents": 11, "elapsed_ms": 9800, "index": [{ "type": "paystub", "pages": "641-658" }] }
One pay stub
Pay stub
type: paystub
confidence: 0.99
Time to classify: ~0.2s
0s — 1s
World-class document intelligence models, served through one API. From raw documents to structured data in seconds.
One API for the whole pipeline
POST /parse
Text, layout and tables from every page, scanned or digital.
POST /classify
A document type for each page, with a confidence score.
POST /split
Split a package into documents and index them to the loan.
POST /extract
Typed fields from your schema, each with its source page.
Built around how a loan file actually moves.
Three building blocks for putting document AI on live borrower files, and each one shows its work.
Send a loan package. Get back an indexed, extracted loan file.
One call parses every page, classifies it, splits the package into documents, indexes them to the loan and extracts the fields you asked for.
Parse
Classify
Split
Index
Extract
Jordan A. Reyes
99%
$64,212.18
97%
2026-09-30
95%
84-•••612
61%
Source: pay stub p.2 · employer_ein below your confidence threshold
Every value comes with its source.
Each field returns with its page number and exact position on the page, so your own app can show where a number came from without anyone paging through a PDF.
Give your agents loan data they can cite.
One tool call returns typed values, the page behind each one and a plain yes-or-no on whether a person should look, so an agent knows when to keep going and when to hand off.
// agent tool call
read_loan_file("loan_4471.pdf")
// returns
{ "monthly_income": 5350.00,
"source": "paystub p.1",
"ask_a_person": false }
Lending teams trust Addy AI
“Indexing loan documents used to take us hours. With Addy’s AI models, we’ve brought that down to minutes, and the results are more accurate and reliable. We can give clients feedback much sooner, which gives us a real edge when we’re competing for their business.”
Amit Pall
EVP, Business Operations, Newfi
Illustrative testimonial
“One API instead of three vendors. Parse, classify and extract come back in the same JSON shape, so we shipped the integration in a sprint.”
Daniel Ortiz · Head of Platform, Wrenfield Bank
Enterprise grade security
SOC 2 Type II & GLBA
Independently audited, built for GLBA safeguards.
Secure access controls
RBAC and scoped API keys.
Zero data retention
Files are deleted once the response is returned.
Get started in minutes.
Try it free on your toughest loan files, like scanned tax returns and 1,000-page closing packages, and see the results for yourself.
Get started for free
