How Rubil handles your data
Rubil is a voice dictation tool. You speak, we format, the text appears in your app. Here's exactly what happens to your data at every step — and what doesn't happen.
What happens when you dictate
- 1
Step 1 — You speak
Your microphone captures audio in your browser (Chrome extension) or on your Mac (desktop app). Audio stays in local memory.
- 2
Step 2 — We transcribe and format
Audio is sent to Groq for transcription, then the transcript is sent to OpenAI for formatting. Both transmissions use encrypted HTTPS. Neither Groq nor OpenAI stores your data — both process transiently and discard.
- 3
Step 3 — Text appears
The formatted text is inserted directly into your app (Gmail, Slack, Notion, etc.).
What we store
| Data | Stored? | Where | Retention |
|---|---|---|---|
| Audio recordings | Never | — | Discarded after transcription |
| Transcripts | Never | — | Discarded after formatting |
| Formatted text | Never on servers. In-memory on Mac app (last 50, cleared on quit) | Local device only | Session only |
| Glossary entries | Yes | Supabase (encrypted in transit, RLS-enforced) | Until you delete them |
| Account info | Yes | Supabase | Until you delete your account |
| Usage counts | Yes | Supabase | Aggregate only, no content |
| Anonymous analytics (Mac) | Yes | PostHog | No transcripts, no content, no IP addresses |
Our data processors
Groq — Transcription
groq.comReceives audio, returns text. Zero data retention. Based in the US.
OpenAI — Formatting
openai.comReceives transcript text + glossary context, returns formatted text. API usage with data retention disabled. Based in the US.
Supabase — Authentication, glossary storage, usage tracking
supabase.comRow-level security enforced. Based in the US (AWS).
PostHog — Anonymous product analytics for the Mac app
posthog.comReceives event names and metadata only (app launched, recording completed, etc.). Never receives transcripts, formatted text, glossary content, or IP addresses. Users can opt out in Settings. Based in the US.
Google — OAuth authentication only
google.comWe use Google Sign-In. Google does not receive any dictation data.
Vercel — Hosts the Rubil backend (API routes) and website
vercel.comProcesses API requests transiently. No persistent storage of user content. Based in the US.
What we will never do
- Never store your audio. Recordings are processed in memory and discarded. There are no voice files on our servers, ever.
- Never store your transcripts. The text version of what you said is used for formatting and then discarded. We don't keep transcript logs.
- Never use your data to train AI. Your dictations are not used to fine-tune, train, or improve any AI model — ours or anyone else’s.
- Never sell your data. We have no advertising, no data brokerage, no third-party data sharing beyond the processors named above.
- Never track what you type. We don't capture URLs, page content, window titles, or anything from the apps you use. The only context we read is which platform you're on (Gmail vs Slack vs Notion) so we can format correctly.
Questions?
If you have questions about how Rubil handles your data, email [email protected]. We'll give you a straight answer.