The promise
Your meeting audio is never saved as a file.
Everything derived from it — transcripts, summaries, notes — lives where you decide: encrypted on our servers with zero-knowledge, on your machine only, or both.
Your choice.
What happens to your audio
No audio file — no .wav, .mp3, or .webm — is ever written to disk. Ever. During recording, audio is held in memory; to survive a browser crash in a long meeting, short batches live for a couple of minutes in your browser's encrypted local database and erase themselves the moment they're transcribed. By the end of the meeting, no audio remains anywhere.
This isn't a marketing claim. The exact function that discards your audio is in our open source code.
See the function on GitHub →Zero-knowledge sync
When you sync, your notes and summaries are encrypted before they leave your device.
Argon2id
Your 12 recovery words become your encryption key through Argon2id, a deliberately slow key-derivation function. This is exactly why, if you forget your passphrase, no one — including us — can recover it.
XChaCha20-Poly1305
Your data is sealed with XChaCha20-Poly1305, the same authenticated-encryption primitive used by WhatsApp and Signal.
Verify in code →Client-side
All of this happens on your device, before anything reaches our servers.
What we store is encrypted data we cannot read. Not because we choose not to — because we mathematically can't.
Three transcription modes
You choose per meeting.
Standard
Transcribed on your machine, summarized in the EU cloud. Audio never leaves your computer.
Local Only
Transcription and summary both on your machine. Nothing leaves, in any mode.
Cloud Fast
Transcribed through Mistral's EU servers in Paris. Audio encrypted in transit, never saved. Fast on any hardware. Pro and Pro Unlimited.
| Your hardware | Standard (local) | Cloud Fast |
|---|---|---|
| MacBook (M-series) | 30 min → 3–6 min | 30 min → 1–2 min |
| Windows / Linux, dedicated GPU | 30 min → 5–10 min | 30 min → 1–2 min |
| Windows / Linux, integrated Intel GPU | 30 min → 50–60 min | 30 min → 1–2 min |
Cloud LLM: Mistral, EU, Zero Data Retention
Our cloud LLM processes only transcribed text — never audio. It runs on EU-hosted Mistral infrastructure, with Zero Data Retention active for our organization and a signed Data Processing Agreement.
Request the DPA reference (PDF) →Our subprocessors
The full list, in plain language.
| Provider | Role | Location | Data |
|---|---|---|---|
| Cloudflare | CDN, DNS & Workers | USA | No user data processed |
| Mistral AI | Cloud LLM & Cloud Fast transcription | EU-hosted | ZDR active, DPA signed |
| Resend | Transactional email | USA | Email address only |
| Polar | Payment processing | USA (Delaware) | Billing data only |
| MailerLite | Newsletter | USA | Email address only (opt-in) |
Where a subprocessor is US-based, it never processes readable meeting content — only the limited data listed. Your meetings stay on your device or are encrypted so that neither we nor they can read them.
Verify it yourself
Don't take our word for it.
- Start a recording in Standard or Local Only mode.
- Open your browser's DevTools — Network tab.
- Record for 30 seconds and watch the outbound requests. In Local Only, nothing leaves your device at all; in Standard, only text — the transcription result — never audio. (Cloud Fast is different by design: there, encrypted audio is sent to our EU servers to be transcribed, encrypted in transit and never saved. That's the one mode where seeing audio leave is expected.)
We are early
We're a small, independent product. We don't publish aggregate transparency metrics yet — our user base is still too small for the numbers to mean anything. We will, when they do. Until then, our code is the proof.