1
Before you start: deployment and accounts
On-premises: installed on your own servers — suited to law firms and organisations with strict data-location rules.
Guide
2
Setting rules: global and grouped
Set rules globally or per group — contract numbers, signing dates, company names.
3
Uploading a file and choosing a rule set
Upload: common office documents, with the rule set that fits (e.g. "shareholder documents").
4
Reviewing flags: remove, add, custom replacements
Review: remove false hits, add missed ones, set replacement text, then publish.
5
Publish, withdraw and export to Word
Publishing locks the version; you can withdraw it at any time.
6
Connecting your AI assistant (one-click MCP setup)
The workbench connects to MCP-capable AI assistants, set up in one click.
One-click MCP setup
Connect connected
Setup one-click setup
One-click setup
7
Using published files in chat
Ask in chat and the assistant fetches the published, redacted file to summarise, compare or draft.
Using published files in chat
Published file: shareholder-special-resolution.docx
the assistant reads the published, redacted version
summarise, compare or draft
8
Good practice: trial with fictional files, review rules regularly
For a trial, we'll tell you in writing where it runs; we recommend testing with fictional files only.
Good practice
Trial with fictional files only fictional files only
Review rules regularly global and grouped rules
Go through the workbench first workbench first
9
FAQ and limitations
Automatic flags can miss or over-flag, which is why every file needs human sign-off before it's published.
Does the original leave my environment?
In a production deployment, the workbench and detection service run on your servers or chosen cloud, and originals are processed there. AI assistants read only redacted versions. We'll state where any trial environment runs up front.
How accurate is detection?
Automatic flags can miss or over-flag, which is why every file needs human sign-off before it's published.
What if staff skip the workbench?
It doesn't intercept all AI use; staff need to put files through the workbench. Pair it with your internal AI policy.
Can AI-edited files be restored with the original details?
This is still being validated; we can test it together in a trial.
Which AI assistants work with it?
Any AI assistant that supports MCP.
Can a team share it?
Team sharing and permissions are on the roadmap; ask us for the latest.
How do we start a trial?
Book a demo. We'll walk through the full flow on fictional shareholder documents, then agree a trial scope.
See it on a fictional version of your own document
A 30-minute walkthrough, from upload to review to AI.