Privacy.

mmumble is a voice keyboard. You talk, we transcribe and clean the text, and it shows up at the cursor. This page explains what passes through our servers, what gets kept, and what doesn't.

The short version: we do not store your audio, your transcripts, or your cleaned text. There is no mmumble account. We don't log keystrokes. The only thing we keep that's identifiable is your email address, and only if you put it in our waitlist form.

What gets sent to our servers

When you tap the mic and dictate, the companion app uploads a short audio recording (AAC in an M4A container) to our backend running on Cloudflare Workers. The backend forwards the audio to Groq for speech-to-text, sends the resulting transcript to Anthropic for cleanup, and returns the cleaned text to your device. The audio and the text live in Worker memory only for the duration of the request — neither is written to disk, a database, or a log.

Each request is signed by a per-device key generated through Apple's App Attest. We see an Apple-issued credential identifier, but no Apple ID, email, name, or device serial number. The credential identifier is used to rate-limit abuse and nothing else; it is not linked to any other profile.

What stays on your device

Your personal vocabulary list — the words you teach mmumble to spell correctly — is stored on your device. It is sent with each transcription request to bias the model toward your spellings, then discarded server-side along with the request. It is never persisted on our backend.

The keyboard extension does not read or transmit anything you didn't explicitly dictate. We do not see what you type with the system keyboard, what app you're in, what's on your screen, or any text context outside the small snippet immediately adjacent to the cursor that iOS hands to keyboard extensions for grammar purposes.

Analytics

The companion app and this website send anonymous product-analytics events to PostHog (e.g. "user opened settings", "transcription succeeded", "visited privacy page"). These events do not include your audio, your transcripts, your vocabulary, your contacts, or your email. App events are tied to a randomly-generated install identifier; website events are tied to an in-memory session identifier that is not persisted across page loads. You can opt out from the app's settings.

Usage tracking

To support the free-tier weekly word allowance, we record how many words mmumble has cleaned for your device in the current week. The record is keyed by the same App Attest credential identifier described above — no Apple ID, name, email, or device serial — and contains only the credential id, the week-start date, the running word count, and a last-updated timestamp. It does not contain your audio, your transcripts, your cleaned text, or which apps you used mmumble in. The record is stored on Supabase (managed Postgres). Rows roll over weekly and we delete entries older than 13 weeks on a periodic schedule.

Waitlist email

If you submit your email on the waitlist form, we keep it until we email you once when the cohort opens, or until you ask us to delete it — whichever comes first. We do not sell it, share it, or add it to a marketing list. The form is powered by Tally; your email is transmitted to and stored by Tally on our behalf while we hold it.

Third parties that touch your data

• Cloudflare — hosts our backend (Workers, KV, Durable Objects) and this site (Pages). Standard request logging applies; audio and transcripts are not in those logs.
• Groq — receives audio for speech-to-text. Per their policy, audio is not retained for model training on the API tier we use.
• Anthropic — receives the transcript and any reference context for cleanup. Per their commercial API terms, inputs are not used to train models and are retained only for abuse monitoring (currently 30 days).
• Apple — App Attest issues the per-device key used to sign requests. Apple does not see request contents.
• PostHog — receives anonymous product-analytics events from the app and this website. No audio or transcripts.
• Tally — powers the waitlist sign-up form on this site. If you submit your email, Tally receives and stores it on our behalf. Their privacy policy governs that processing.

What we do not do

• No advertising. No ad SDKs, no ad identifiers, no profile sales.
• No location tracking.
• No reading your contacts, photos, calendar, or other apps.
• No keystroke logging from the keyboard extension.
• No training of any model on your voice or text.

Children

mmumble is not directed at children under 13. We do not knowingly collect data from anyone under 13. If you believe a child has used the waitlist form, email us and we'll delete the record.

Your rights

Because we don't keep transcripts or audio, there is nothing of that kind to export or delete. For the limited data we do hold — waitlist emails, anonymous analytics — you can ask us to delete it at any time. Email [email protected] and we'll handle it within 30 days.

If you're in the EU/UK, the legal basis for the small amount of data we process is legitimate interest (running the service and preventing abuse) or consent (waitlist email, opt-in analytics).

Changes

If we change this policy, we'll update the effective date at the top and, for material changes, surface a notice in the app the next time you open it. We won't quietly broaden what we collect.

Contact

Questions, deletion requests, or anything else: [email protected].