Why I made this: I want to talk to AI (voice or text) without my conversations being logged, used for training etc.
Requires an OpenRouter account. No subscription: pay-as-you-go with most conversations costing a fraction of a cent (a couple of cents for requests with search enabled)
Privacy:
- Uses OpenRouter to route to Gemini 3 Flash on Google Vertex with Zero Data Retention (ZDR).
- Prompts and responses aren't stored or logged at any layer.
- OpenRouter keeps only metadata like timestamps and token counts.
How it works:
- Single HTML file – download and self-host, or use the hosted version https://martinpllu.github.io/spraff
- Log in to OpenRouter via OAuth (API key kept in browser local storage until logout)
- Audio recorded in browser, sent directly to Gemini for processing (no separate transcription step)
- Text-to-speech via your device's local voices
Technical notes:
- Gemini 3 Flash is currently the only model that supports both audio input and ZDR
- Provider is pinned to Google Vertex with zdr: true in the API request
- No build step, no dependencies – just one HTML file
- Current status: beta, tested on MacOS, Android and iOS. Should work but issues/feedback welcome.