Skip to content

How it works

Get started in 6 steps.
You do the first 4 sitting down.

There is no implementation project here. You create an account, tell us about your business, drag your documents in, and connect the AI account you already use. The governance happens underneath, on every upload, forever.

  1. 01

    Create your account

    Name, email, password. The usual.

  2. 02

    Sign in

    Your own workspace is waiting, already set up for your industry.

  3. 03

    Fill in your business details

    Company name, address, phone, license and insurance numbers, department leads. These become the facts your assistant answers from when no document covers a question.

  4. 04

    Upload your documents

    Drag them in. Each one is read, sorted, versioned, and checked for sensitive information before it is stored.

  5. 05

    Connect your AI

    Tell us which account you use, copy one link and one block of instructions, paste them into your project. That is the whole connection.

  6. 06

    Ask it anything

    From that moment your assistant answers from your company’s documents, and keeps up as they change.

Step 03 · Business details

The facts your assistant answers from when no document covers the question.

Your address, hours, license and insurance numbers, and who runs each department. Fill in what you know and save — you can come back for the rest. Only the essentials hold anything up.

The Business details screen with fields for company address, email, hours, legal name, phone, website, languages and year established.
54 details in the full form, 16 of them essential. Phone numbers, emails and web addresses are reformatted to one standard form on save — exactly what the AI will quote.

Step 04 · Upload

Choose the file. We handle the rest.

Pick the department and what kind of document it is — or let the system read the file and propose a filing for you to confirm. It works out the name, the version number, and which document this one replaces.

The upload form with fields for file, department, document type and title, and a choice between new document and new version.

Before you publish

If a document of that kind already exists, the form tells you before you publish: this is a new version, and it will take over from the current one. Nothing is overwritten by surprise, and nothing is ever deleted.

Word, PDF, Excel, CSV or plain text. Screened for personal and health information before anything is stored.

Accepted formats

What happens on upload — in this order

  1. 01

    Read

    The text is read out of the file you dropped in, before anything is kept.

  2. 02

    Screen

    Sensitive-data detection runs on that text. A file that fails is refused, not stored.

  3. 03

    Sort

    The department and the kind of document are worked out against your industry’s list.

  4. 04

    Version

    The document gets its permanent identity, and this file gets its version number.

  5. 05

    Publish

    It becomes the version that counts, under a name that never changes shape.

  6. 06

    Retire

    The one it replaces is kept and dated, but is no longer answered from.

  7. 07

    Update

    Your assistant picks it up from that moment. Nothing to re-upload, nobody to notify.

  8. 08

    Record

    Who did it, when, and what it replaced — written down and never edited.

The screen

Refused before it is stored, not flagged after.

Sensitive-data detection runs on the text before anything is kept. A document carrying data we don’t accept never reaches your system, and never reaches the assistant.

  • Social Security Numbers
  • Payment Card Numbers Luhn-validated
  • Protected Health Information (PHI)
  • Medical Record Numbers (MRN)
  • Bank & Routing Numbers
  • Scanned Files With No Text Layer images, audio, etc.
The activity record showing documents published alongside uploads refused by the screen, each naming the category that triggered the refusal.
Refusals are recorded like everything else — the file name, the reason, and who tried.

Step 05 · Connect your AI

One link and one block of instructions. That is the whole connection.

Tell us which assistant you use — Claude or ChatGPT — and which Google account will read your documents. We write the instructions for you from your live setup, so they can never describe a system you no longer have.

The AI configuration screen: choosing which assistant you use, and entering the Google account that will be granted read access, with the link to copy.
The generated project instructions block, ready to copy into a Claude or ChatGPT project.
Written for you from your live setup. Paste it into your project once; you never paste it again.

Your documents stay yours

They live in storage your own company owns, under your own account. We never hold the files. Leave, and every one of them is exactly where it already was.

One connection, not a hundred

Your project gets one thing to point at, and that one thing always knows which version of each document is current. That is why nothing has to be re-uploaded.

Your own house rules

Answer in English and Spanish. Never quote pricing outside the company. Always name the document. Your instructions survive every regeneration.

Day to day

Updated document? Upload the new version. That’s it.

No retraining. No rebuilding. No announcement to the team. Because your project reads what is live rather than copies of your files, the answer changes the moment the document does — in the same thread, on the same question.

Step 01

You upload the revision

The same document you already keep — a price list, a warranty, a field manual.

Step 02

It becomes current

What you just uploaded becomes the version that counts. The one before it is kept and dated, but no longer answered from.

Step 03

Answers follow

From that moment every answer uses the new one. Ask the same question again and watch it change.

The Recently published list showing the last six documents to enter the corpus with their doc codes, versions and who published them.

Nothing is deleted

Every version you have ever published is still there, dated, with a record of who replaced what and when. When a dispute lands, you can show exactly what you sent in March — and prove it wasn’t what you send today.

The guardrails

What it will not do.

Governance that depends on a model behaving well is not governance. Most of this is structural — the superseded file is not somewhere the assistant can reach, whatever it decides to do.

  • It will not answer from a retired version

    Once a document has been replaced, it stops being something your assistant can answer from. Being able to open an old file is never evidence that it is current.

  • It will not invent policy

    Where your documents and your business details do not give a clear, authoritative answer on policy, legal, compliance or safety, it says what is and is not covered rather than filling the gap.

  • It will not treat a placeholder as fact

    [Company Name], [Project Name] and sample names in templates are filled from your recorded business details, never answered as if they were real data.

  • It will not silently drift

    The instructions your assistant reads are rewritten from your live setup every time it changes. They cannot describe a department list you no longer have.

It also catches what a proofreader misses. In testing it flagged a dead web address in a client’s own marketing document, and traced it back through two revisions.

Before you start

What you need on hand.

Your business facts

Company name and legal name, address, phone, hours, license and insurance numbers, and the person who runs each department.

Your documents

Word, PDF, Excel, CSV or plain text — filed against 471 document types across 51 departments. Whatever you already keep.

A Google account

The account your assistant will sign in with. If your company runs on Microsoft 365, a Google account on your existing work address is enough — your email stays exactly where it is.

On one AI system

No IT project. No consultant.
Nobody has to learn a new system.

They just ask questions — and the answer comes from the document that is current today.