Show HN: Spraff – Voice and text AI chat, self-hostable, no data retention (github.com)

1 points by pllu ↗ HN
Hi HN! I built a simple, self-hostable AI chat app that focuses on voice input and privacy.

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.

Code: https://github.com/martinpllu/spraff

0 comments

[ 0.30 ms ] story [ 9.6 ms ] thread

No comments yet.