Trust & transparency
Where your data goes
Minuta uses a cloud AI chain for transcription, summaries, and copilot answers. This page documents exactly which data leaves your computer, which providers handle it, and what we never do.
What stays on your machine
The following never leave your computer unless you explicitly export or share them:
- Audio capture — Minuta records system audio locally on the desktop app.
- Transcript database — every transcript is stored in a local SQLite database on your computer.
- Meeting summaries — generated summaries are saved locally, not server-side.
- Copilot answers — suggested answers during the meeting are stored in the local database.
- Exports — any file you export (text, PDF, etc.) stays on your machine until you choose to send it.
- Attached context documents — files you attach to a meeting for copilot grounding stay local.
What we send to the cloud, and why
To deliver transcription, summaries, speaker identification, copilot answers, and optional translation, Minuta routes narrow payloads through a reviewed multi-provider chain. Audio is sent only for transcription and diarization; text payloads go to summaries and copilot. We do not log these payloads on Minuta's API.
Transcription
Deepgram (primary ASR)
OpenAI Whisper (fallback)
Groq Whisper (last resort)
Summaries & copilot
OpenAI gpt-4.1-mini (primary)
Anthropic Claude Haiku 4.5 (fallback)
Groq (last resort)
Speaker diarization
Modal / pyannote — staged audio uploaded to a private short-lived S3 object
Translation (optional)
DeepL — transcript text only, never audio
Subprocessors
All subprocessors are listed here and in our Privacy Policy. We update this table whenever the provider chain changes.
| Provider | Purpose | Retention note |
|---|---|---|
| Deepgram | Automatic speech recognition (transcription, primary) | No audio retained per Deepgram's zero-retention API contract |
| OpenAI | Transcription fallback (Whisper); summaries and copilot answers (primary) | Transient processing; data not used for training by default under API terms |
| Anthropic | Summaries and copilot answers (fallback) | Transient processing; data not used for training under API terms |
| Groq | Transcription and LLM inference (last-resort fallback) | Transient processing per Groq API terms |
| Modal / pyannote | Speaker diarization — identifies who spoke when | Audio staging object deleted after diarization completes |
| AWS (S3, Lambda, API Gateway, Cognito, DynamoDB, CloudFront) | API hosting, auth, account metadata, short-lived audio staging, CDN | Account metadata retained while account is active; staging objects deleted after use |
| DeepL | Real-time translation (optional, when enabled) | Text only; not retained per DeepL API terms |
| Stripe | Payment processing and subscription management | Stripe holds card data; we store only the Stripe customer ID and subscription state |
| Google reCAPTCHA v3 | Anti-bot on site forms | Score returned; not persisted by us |
| Sentry (optional) | Error diagnostics — never includes meeting content | Stack traces and metadata; gated on VITE_SENTRY_DSN env var |
| Plausible (optional) | Anonymous pageview analytics — no cookies, no meeting content | Aggregate stats; gated on VITE_PLAUSIBLE_DOMAIN env var |
What we do not do
- We do not train AI models on your meeting audio, transcripts, or copilot answers.
- We do not retain audio beyond the immediate API call — audio is sent for transcription or diarization and not stored on our servers.
- We do not use a meeting bot — Minuta joins no meeting as a participant; it captures audio locally on your computer.
- We do not sell or rent your meeting content to third parties, data brokers, or advertisers.
- We do not log raw audio, transcripts, prompts, or generated responses at the Minuta API layer.
- We do not collect desktop analytics unless you opt in, and opt-in events never carry meeting content.
Have a question?
If you have questions about how Minuta handles your data, reach us directly.