Zero data retention agreements are "trust, don't verify."
That's why we built Private Mode: your prompt is encrypted on your device to an attested enclave. We can't read it, and you can check that yourself.
Filtering by: privatemode · clear filter
Web search is now optional in NanoGPT Private Mode, including the local proxy.
NanoGPT cannot read queries or results. Exa receives queries; Cloudflare processes fetched URLs and page contents outside the inference enclave. Search is not end-to-end confidential from those services.
Off by default. Additional search fees apply, plus model token charges.
Details:
Private Mode fixes are live on NanoGPT.
Fewer intermittent failures with Kimi K3 and Gemma 4, longer timeouts for reasoning, and hiding reasoning no longer changes whether the model thinks.
Local proxy users: update to 0.2.14 or later and restart.
Private Mode now supports image attachments with Kimi K3 and Gemma 4 31B.
Upload JPG, PNG, GIF, or WEBP images for private visual analysis. Your browser places the image directly inside the encrypted request sent to the model's trusted execution environment.
The image is never uploaded to NanoGPT object storage or sent to separate image-description or OCR services.