Security

What stays on your computer.

What is sent to the AIs, what Kaviar processes, and what never leaves your computer.

Redaction happens on your computer

Detection and replacement run locally, on your computer, so your sensitive details never leave it. Neither the AI nor Kaviar ever receives them.

The AI receives made-up equivalents

Each detected sensitive detail is replaced with a plausible value; the rest of the message is sent as is. The originals are put back in the answer, on your computer.

Your content stays on your computer

Conversations, routines, Context, Vault and history live in the app, on your computer, and Kaviar never sees their content. On Kaviar’s side, only your account and content-free usage metrics are processed.

Designed for GDPR

Kaviar only processes your account information and content-free usage metrics. The detail, purpose by purpose, is in our privacy policy.

Your sensitive data stays out of training

Your sensitive details are never part of the message, so a provider’s training policy cannot reach them. You can check this at any time in the AI view.

Delete everything in one click

Conversations, Vault and history can be deleted right from the app.

Transmission

How your messages reach the AIs

With Kaviar creditsFrom your computer to the AI, through Kaviar’s servers. Only the protected version passes through, and it isn’t stored.
With your own API keysStraight from your computer to the AI, so Kaviar sees none of it.
Either wayYour sensitive details stay on your computer, and you see word for word what was sent to the AI.

What is kept, and where

Your sensitive detailsOn your computer only
Conversations and their historyOn your computer, until you delete them
The mapping between originals and made-up equivalentsOn your computer, with each conversation
Context, Vault and routinesOn your computer. Synced with encryption across your computers with the Pro plan, if you turn it on
The protection log (number of protected details, by type and date)On your computer
Messages sent with Kaviar creditsProtected version only, never stored

Detection

How Kaviar recognizes what’s sensitive

Kaviar uses two detection layers, both on your computer. The first analyzes every message and file instantly as you type, looking for first and last names from every culture, companies, places, IBANs, official numbers, API keys and dozens of other formats, in 7 languages.

The second, an entity-recognition model built into the app, rereads the message right after to catch anything that slipped past the first. The two reads are independent, and no data is sent anywhere to be analyzed.

Did Kaviar miss something? Select it to protect it, or add it to the Vault so it’s protected in every conversation.

Compliance

Built for regulated industries

GDPR. Personal data is replaced on the user’s computer before anything is sent: the AI provider receives none of it. Kaviar puts data minimization into practice without changing how people work.

EU AI Act. Kaviar is not an automated decision-making system: it only controls what information reaches AI models, and you stay in charge of every message.