How Does PrivatAI Keep My Data Safe?
The Principle: Less Is More
It's simple! We keep your data safe by not storing it in the first place. Unlike other AI providers, your data only exists for the few seconds it takes to compute an answer. Once the response has been sent, nothing is left behind — there is no conversation archive to search, subpoena or leak.
The Technical Process
Every time you send a message to our chat AI, the following happens:
- Encryption in transit: your browser opens a TLS connection, so nobody on the network path can read the message
- Transmission: the request arrives at our application servers in Germany (Hetzner, Falkenstein)
- Forwarding to the model: the application passes the request on — again over TLS — to the inference data centre in France (Scaleway), where the model itself runs
- Computation: the model generates the answer in memory. Nothing is written to disk at either stop
- Response: the answer travels back the same way, encrypted at every step
- Display: your browser decrypts it and shows it in the chat window
Both stops are inside the EU — the application in Germany, the AI processing in France. No US provider is involved in handling your prompts, so the CLOUD Act has nothing to reach for.
Visual Representation of the Data Flow
Hetzner, Germany"] B -->|🔒 TLS| C["🇫🇷 AI inference
Scaleway, France"] C -.->|🔒 TLS| B B -.->|🔒 TLS| A style A fill:transparent,stroke:#f97316,stroke-width:2px,color:#f4f4f5 !important style B fill:#f97316,stroke:#f97316,stroke-width:2px,color:#f4f4f5 !important style C fill:#f97316,stroke:#f97316,stroke-width:2px,color:#f4f4f5 !important
Every transmission is TLS-encrypted, and neither stop keeps a copy.
One Point of Honesty About Encryption
This is transport encryption, not end-to-end encryption — and the difference matters. A model cannot compute an answer on text it is unable to read, so your message is necessarily in plaintext in memory for the moment it is being processed. Anyone who promises you end-to-end encrypted AI is either running the model on your own device or being careless with the term.
What we can promise instead, and do: your message is never written to disk, never logged, and never used to train a model.
What Does This Mean for You?
Your message flows through without leaving traces — no content logs, no training on your data, no body of data that could be caught up in an incident.
There is exactly one exception, and we would rather state it than let you find it in the fine print: if malicious activity or a disruption to service quality is detected, the inference data centre may temporarily retain the content of individual requests in order to investigate the cause. Even then it is never used for training and never accessible to third parties. The full picture is on our security page.
Important: You are responsible for saving the results from your conversation yourself. They disappear forever as soon as you close your browser tab or start a new chat.
