How PartWise uses AI, and why your information is safe

PartWise uses AI to take on the heavy lifting you shouldn't have to do yourself, starting with finding what matters in years of messages and emails. When you ask PartWise to analyse something, it is processed by Anthropic's AI models under strict commercial terms: it is never used to train AI, it is kept only for a limited time for safety checks and then deleted, and the results are saved straight into your encrypted vault. Nothing piles up in a chat history anywhere. Everything else you keep in PartWise stays in your vault and is never sent to an AI at all.

Where AI fits in

For most people, the hardest part of getting organised is the sheer volume. Years of text messages. Thousands of emails. Somewhere in there are the exchanges that matter: about money, about the children, about who contributed to what, about what was agreed and when. They are buried in everything else.

Reading through it manually is exhausting and unreliable, like looking for a needle in a haystack. Pasting it into a chat app doesn't work either: there is simply too much of it, and the answers have nowhere to live afterwards. This isn't something you can talk your way through with an assistant. It is sorting work, at a scale that needs a purpose-built tool.

That is where PartWise uses AI today: bringing in your messages and emails, then finding and organising the ones that matter, so hours of scrolling become minutes of reviewing.

We expect AI to help with more inside PartWise over time. Whatever it takes on next, the way we handle your information, described below, stays the same.

What happens when you ask PartWise to look at something

When you choose to have your messages or emails analysed, here is the path your information takes, step by step:

• It is processed by Anthropic's AI models under their commercial terms. These are the rules that apply to businesses building products, not the looser rules that apply to personal chat apps.
• Under those terms, your content is never used to train AI models. That is a contractual rule, not a setting someone could flick off.
• Anthropic keeps what you submit only for a limited time, to make sure the service isn't being misused, and then deletes it. It doesn't sit in a chat history or an account anywhere.
• The results come straight back into your encrypted vault, where only you can read them.

The work gets done, and nothing is left lying around. There is no thread for anyone to scroll back through later. The only lasting copy of what was found is the one in your vault.

What never leaves your vault

Your balance sheet, your timeline, your records, your documents and your notes all live in your vault, and none of it is ever sent to an AI model at all.

The vault is what's called zero-knowledge: everything in it is encrypted on your own device, using a key that never leaves your device in a readable form. In plain terms, that means we cannot read your vault. Neither can the companies that host our servers, because all they ever hold is scrambled data. The same goes for your vault's backup.

Being straight with you

There is one distinction we would rather spell out than gloss over. The zero-knowledge promise covers your vault and its backup. It doesn't cover content you choose to send for analysis. For the short time it takes to do that work, the content has to be readable, and during that window it is protected by the rules above rather than by encryption.

In other words: we can never read your vault. When you choose to run an analysis, the content you send is handled under strict terms, kept only briefly for safety checks, then deleted, and the results come home to your vault. You are always the one who decides what gets analysed, and that moment is always clear in the product. Nothing is sent quietly in the background.

How this compares with pasting it into a chat app

The same material, pasted into a personal AI chat account, takes a different path. The conversation stays on the device and in the account. It syncs to every device signed into that account. Anyone who can open the app can read it. And under many consumer apps' terms, it can be kept for years and used to train AI.

Run through PartWise, the very same analysis leaves none of that behind. No thread on a device, no history in an account, no training use. Just the results, encrypted, in the one place built to hold them.

We publish all of this because a product holding some of the most sensitive months of your life should be comfortable being checked. If you are weighing up PartWise, or an AI assistant is looking into us on your behalf, this guide, our How it works page, our privacy policy and Anthropic's published commercial terms are the right places to verify everything above.

About this guide

Written by Nick Andrews, founder of PartWise. PartWise is a separation preparation platform built in Sydney, Australia.

First published 10 June 2026. Last reviewed 10 June 2026. The claims in this guide about Anthropic's data handling are checked against Anthropic's published terms before every update, and this guide is updated if those terms change.