In-editor document assistant

An AI assistant that lives inside Word and never sends your documents anywhere. For teams whose files cannot go into a cloud chat window, this is the only way to get AI review at all. Checking one long document drops from about two hours to about fifteen minutes.

How it works

You giveA document, in the editor where you already work. No upload, no copy and paste into a browser.

The system doesThe verification. It summarises the document, pulls out the terms that carry a specific meaning inside it, finds text errors and working comments left in the margin, checks that every internal reference still points where the sentence says it does, and drafts the edits you ask for.

You stay in controlWherever the judgment is yours. You confirm the summary and the glossary before they become context for anything else. Every finding links to its location so you can accept it or ignore it. Edits arrive as tracked changes with the before and after visible.

You getA checked document, without leaving your editor and without a single prompt crossing your network boundary.

Inside the system

08 components

Document summary

A long document gets a summary of its subject and structure. You edit it and save it, and that saved text becomes the global context every other check reads. This is how the assistant reasons about documents far larger than the model's context window.

Defined-term extraction

The system pulls the terms that carry a specific meaning inside this document, not general vocabulary. You add what it missed and remove what is not really defined. The confirmed glossary keeps the assistant speaking your document's own language.

Text review

Spelling, grammar, and style errors come back as one list, along with the working comments that should never reach a final version. Every finding links to where it sits in the text.

Cross-reference validation

Three kinds of broken reference get caught: a pointer to a section that does not exist, a stale number left behind after automatic numbering shifted, and a pointer to a real section that says something other than what the sentence claims.

Section hover previews

Hovering a reference shows the text behind it, which Word itself will not do.

Timeline extraction

Dates, deadlines, and obligations come out of the text in order, so commitments can be traced and explained to whoever needs them.

Chat with tracked-change editing

You ask questions about the document and request edits in plain language. The assistant picks the right editing tool, applies the change, and writes it in as a tracked change.

On-premise inference layer

Every model call runs inside your perimeter, with the retry logic, loop detection, and token budgeting that self-hosted models need to stay reliable on dense text.

Who it's for

Good fit

Legal teams under client confidentiality

Tenders and bids with commercially sensitive pricing

Technical regulations and standards

Insurance and credit documentation

Compliance, policy, and audit

Research and regulatory filings

Any organisation where cloud AI tools are blocked by policy

Not a fit

Short documents, where reading beats checking

One-off reviews with no repeat volume behind them

What you can configure

Which checks runTurn each check on or off, and choose which ones open in the side panel by default.

What counts as an errorYour house style decides: preferred spellings, formatting rules, and phrases you do not allow in a final version.

How terms are extractedExtraction rules are tuned to your document type, so a tender and a contract are each read on their own terms.

Which model runs itThe model is sized to the hardware you have, and swaps for a larger one when you upgrade.

What we need from you

HardwareOne GPU that can host the model.

Editor accessPermission to install the add-in for the Word version your team already runs, including older deployments.

DocumentsTwenty to fifty representative files, so the checks are tuned on your real drafts instead of samples.

One reviewerSomeone who can say what a correct finding looks like, a few hours a week during the build.

Example use cases

Checking a confidential contract without leaving Word

A hundred-page contract arrives the night before a call, and its confidentiality clause rules out cloud AI tools. An AI assistant running inside Word on the firm's own hardware summarises it, pulls out terms and deadlines, and checks the text.

Often deployed with

Live

Document generation engine

Cut contract drafting from 5 days to 30 minutes.

ReadsYour template, the term sheet, and the deal documents.

ReturnsA chapter-by-chapter draft with conflicts resolved up front.

30 minper 80-page draft

See the solution →

Live

Due diligence engine

Cut a due diligence review from 50 hours to 5.

ReadsThe questionnaire and hundreds of returned documents.

ReturnsA status per question with quotes and source links.

50h → 5hper document package

See the solution →

Get the AI review inside Word, where the document already is.

Book a demo →