Private document AI · with receipts

Ask anything.
Leak nothing.

Contracts, handbooks, manuals, minutes, images, audio — SharpSignal turns the documents your work already runs on into a system that answers questions with the page it got the answer from. It runs on a machine you own, so the sharpest thing in the building never phones home.

Drop in the mess. Ask like a human.

What it does, plainly: you point it at the documents your work runs on, then ask questions in plain language. It answers in a few sentences and names the documents behind the answer, so you can open them and check it yourself.

Underneath it hunts by meaning as well as wording — several searches at once, results pooled — so a question phrased nothing like the document still finds it. And it may only cite what the search turned up. No passage found, no claim made.

Three promises, and we'll be held to them.

Cited, or refused.

Every answer names the documents it came from. When your documents don't support an answer, it says so instead of guessing — a refusal you can act on beats a confident paragraph you have to verify.

Runs on your hardware.

The server sits in your office; documents and questions never leave it, and no outside service sees either. If we disappeared tomorrow, it would keep answering.

Receipts, not benchmarks.

We write our measurements down — including the failures — and they go public with the toolchain, so what this system can and can't do is checkable before you pay for anything.

a question your documents answer + the pageit came from “your documentsdon’t say”

Both outcomes are answers. Only one of them guesses — and it's not ours.

Not everyone sees everything.

Reader permissions apply before the search runs, so an answer is built only from documents that person may read — and every read is written to an access log you keep. Contracts and HR files can sit alongside the handbook.

Real receipts · the questions and the answers, verbatim

Three receipts, and one you can open right now.

a real business document · asked 2026-08-06

We pointed the stack at OSHA's Small Business Safety and Health Handbook — 98 pages, fetched from osha.gov — and asked: “How often do we need to inspect our fire extinguishers, and what has to be recorded?”

The answer, verbatim: “Fire extinguishers are visually inspected monthly, and the inspection is recorded [1]. Additionally, fire extinguishers receive an annual maintenance check, and the maintenance is recorded [1].” — citing page 30, and page 30 says exactly that, word for word. Every read wrote a row in the access log while it happened.

The handbook refusal.

“What does our health insurance plan cover?”“I don't have enough information.” — a question from a different binder. The safety handbook holds no benefits, and the system said so instead of borrowing an answer from the wrong shelf.

The refusal that proves it knows which document it's reading.

The live table.

Ask it yourself. A public deployment anyone can query without an account — browse its rulings, answered and honest abstains alike, at RuleSage's Monopoly table.

Two quieter uses: onboarding — new hires ask the documents instead of your senior staff — and continuity, when someone leaves and what they knew stays answerable instead of going dark.

Why not just use the cloud one?

It can't read your files until you hand them over — every question uploads a document to someone else's computer: someone else's retention policy, someone else's subpoena, someone else's breach. SharpSignal brings the model to the files, not the files to the model.

Want it for your documents?

SharpSignal is installed, not subscribed.

You buy the machine; we install and configure it, load your documents, and test it on questions your people actually ask before you sign off. Same terms for everything strata→signal builds — the deal, at strata→signal →

We run it ourselves.

RuleSage, a public question-answering service anyone can use without an account, is built on SharpSignal and answers from a document library on our own hardware — same code, same citations.

Tell us about your documents →

a short note is plenty — what the documents are, and what you wish they could answer

runs on

used by

proven by

a strata→signal property · hello@strata2signal.com · say hello