Draft an NDA from your own precedents
Point Hexagone AI at your NDA library. Ask Claude for a new template tailored to the prospect; it reasons on the anonymized copy.
The precedents that would make an AI genuinely useful are the ones you cannot share. A generic model drafts a generic NDA; your library holds the clauses you have already negotiated, the carve-outs your firm insists on, and the wording that has survived review. It also holds the name of every client and counterparty you have ever acted for, which is exactly what professional secrecy forbids you from uploading.
How it works
Point at the folder
Give Hexagone AI the directory holding your NDA precedents. It builds a protected mirror on your machine and keeps it in step as you add files.
Add the mirror as a source
In Claude or ChatGPT, add the protected folder as a source. What reaches the model is the masked copy: same clauses, same structure, no identities.
Draft, then restore
Ask for the new agreement. The draft comes back carrying markers; Hexagone AI resolves them to the real names on your machine, so you read and send a finished document.
Generate an NDA for [o.PROSPECT_1] based on the templates in [c.NDA_LIB].
What gets masked here
- Client and counterparty names
- Company names and registration numbers
- Signatory names and job titles
- Addresses and registered offices
- Dates and amounts that identify a deal
What you get back
A first draft built from your own negotiated language rather than a model's idea of an NDA, with the clause structure your firm already stands behind. The precedents never left the building.
How accurate is the detection?Questions on this use case
- Does the AI see which clients I have acted for?
- No. Names, companies and addresses are replaced before anything leaves your machine, and the same party keeps the same placeholder across every document in the folder, so the model can still follow the logic of an agreement without knowing who signed it.
- Will the draft come back full of placeholders?
- Only until it reaches you. The mapping from placeholder to real value is kept locally, so Hexagone AI resolves the markers in the returned draft and you read it in clear text. The version the model saw and the version you read are not the same document.
Try it on this exact task.
One week free, no credit card. Nothing is uploaded, so you can run it on a real file without asking anyone's permission.
macOS Apple Silicon and Windows 10/11