Durham, North Carolina · Private deployment

Your documents.
Your models.
Your vault.

Ledgerlight AI deploys private language models inside your own environment. Invoices, claims and contracts are read on your side of the wall, and a person confirms every field before it reaches a system of record.

Request a pilot See how it works

Invoice & claims extraction Contract review models Human-in-the-loop review API integrations

A single printed page lit by a narrow band of light in a dark room — Fig. 01 — Documents stay inside your perimeter

How it works

The model lives where the documents live.

Nothing goes out to a third party. Inference, retrieval and logging run behind your own access rules, and every value the model returns points back to the page it was read from.

Ingest

Documents arrive from a watched folder, a scanner, or an upload flow your team already uses.

Extract

Structured fields come back with page-level provenance attached to every value the model returns.

Review

A queue holds the fields that need a human decision, so corrections happen before anything is recorded.

Deliver

Finished data is pushed into the finance or claims systems you already run, through an endpoint you control.

A short rack of servers in a small locked room with cool blue status lights — Fig. 02 — A private deployment, on hardware you control

What the model reads

Three document types, in the shape they arrive.

Scans, photographs, exports and the forms nobody designed. Every extraction keeps a link back to the source page, so a reviewer can check the model's work in a click.

Invoices

Header fields, line items, tax and totals, read from the files suppliers actually send, with each value tied back to its source page.

Invoice extractionLine items

Claims

Claim packets arrive as scans, photographs and half-finished forms. The model returns the fields a review team needs and flags the ones it will not guess at.

Claims extractionConfidence flags

Contracts

Clause-level review models surface the terms your team cares about, and leave the judgment call to the person who signs.

Contract review modelsHuman in the loop

Every deployment ships with the review tooling and API integrations that put the output where your team already works.

Shipped work

What is already behind the door.

Engagements are described by their shape, not by client names. The documents and the figures stay where they belong.

A person reviewing scanned documents on a large monitor in a dim office, seen from behind — Fig. 04 — A review queue in production

Insurance · Claims

Claims packets, read end to end

Submitted claim documents go in; the fields a review team needs come out, each one traceable to the page it was read from.

ExtractionHuman review

Finance · Invoices

Invoice intake inside the walls

Supplier invoices are read in the finance team's own environment and handed to their existing systems through an integration they control.

Private deploymentAPI

Contracts · Review

Contract review with a reviewer

A review model marks the clauses that matter, and a person confirms every call before anything is recorded.

Review modelsHuman in the loop

We name clients only with their permission.

Start here

Put your documents behind the door.

We scope pilots around one document type and one team. Bring us the paperwork you would never send to a public model.

Tell us which document type costs your team the most time.
We scope a pilot against your own files, in your own environment.
You review the output before anything reaches a system of record.

Request a pilot

Durham, North Carolina

Your documents.Your models.Your vault.