Skip to content

08 / Guides

What a private AI chat stores, and how to check

Private can mean many things. Five questions that show what an AI chat service really keeps, and our own answers to them.

08-2Published Sep 26, 2026

Almost every AI chat now calls itself private. The word covers very different setups, from encryption with a key per account to nothing more than a promise. Five questions cut through it. Ask them of any service, including us.

1. Where are my chats stored?

Some services keep history only in your browser, so it disappears when you switch devices. Others keep it on their servers. Ask which, and in which country.

nonex: on our servers in the European Union, so your history follows you to every device.

2. Who holds the encryption keys?

"Encrypted" alone says little. Ask which data is encrypted, with what, and what happens to the keys when you leave.

nonex: chats, images and search sources are encrypted at rest with AES-256-GCM, with a separate key for each account. Deleting your account destroys its keys.

3. Where are answers computed, and is anything kept there?

Many chat products forward your messages to another company's model. Each hop is another place your text can be logged.

nonex: one model that we run ourselves. Requests are never routed to other companies. Model servers are rented from independent providers, whose operators can technically access a server while it runs. They do not store chats and are shut down when idle.

4. Are my chats used for training, ads or analytics?

Look for a clear no in the privacy policy, not "we may use data to improve our services".

nonex: no training, no ads, and no third-party analytics or tracking scripts. The app has no analytics at all, and this site sets no cookies.

5. Can I export and delete everything?

You should be able to take your data with you and remove it, including from backups after a stated period.

nonex: export and account deletion are in Settings. Deleted data ages out of backups within 37 days.

The honest summary

No hosted AI can be private from its own operator unless the model runs in hardware the operator cannot inspect. What a service can do is keep less, route less, encrypt what it stores and tell you exactly where the limits are. That is the bar we hold ourselves to.

Try nonex for an hour