Private by design.
Your text stays yours.
Hold the shortcut and speak: on Mac the audio travels encrypted to our server and on to Groq for transcription, and the text is cleaned up the same way — neither request keeps a recording or a transcript. On Windows the same work runs on your PC; on iPhone Apple’s speech recognition does it. What you dictate lands where your cursor was. With Audios sync on (it starts on, after a notice) your Mac dictations and the voice notes you import on iPhone are kept on our servers for 90 days to sync your devices — audio files in the EU — and you can delete them anytime.
Used every day inside the projects we run:
Kept only for what you use, never sold
Transcription and cleanup keep nothing: audio and text are processed and discarded. Your dictionary and history live on your Mac — a few dictionary terms travel with each dictation as a spelling hint for the transcription. What we keep is tied to features you can turn off: Audios sync (90 days; audio files in the European Union on Cloudflare R2, texts on Cloudflare D1), dictations for the AI apps you connect via MCP (up to 365 days) and dictations sent for review (90 days). Turn them off in Settings, or delete every synced audio in one tap.
Check it, don’t trust it
Ask what leaves, and where it goes.
Every hop is written down
On Mac the audio goes to api.hearflow.it and from there to Groq, Inc. (United States) for transcription; text cleanup, summaries and the assistant go through Groq too. Synced audio files are stored in the European Union (Cloudflare R2) and their texts on Cloudflare D1. Every hop is in the Privacy Policy.
Mapped, not hand-waved
Providers, purposes, retention and extra-EU transfers are all listed in the Privacy Policy, with the safeguards of artt. 44-49 GDPR for the ones based outside the EU.
Transcription keeps nothing
Audio and text sent for transcription and cleanup are not stored. What stays is usage counters tied to your account and the content of the features you leave on: Audios sync (90 days), MCP (up to 365 days), dictations sent for review (90 days).
Kept on your disk, or only where you choose.
Each of these is a file you can open or a switch you can flip, and dictation behaves exactly the same with all of them off.
Your dictionary lives on your Mac
The names and terms HearFlow learns live in your user folder and are applied on your computer. Up to 12 of them travel with each dictation as a spelling hint for the transcription and are not stored; the other ways words can leave are listed in the Privacy Policy, each with its switch.
You set the memory
Keep a local history of 90 days (up to 2,000 entries), or choose not to store transcripts at all. Audios sync is its own switch: turn it off, or delete every synced audio from our servers in Settings.
Your voice is not fuel
Transcription keeps no audio, and no audio is ever used for training — synced audio included. Sending dictations for review shares text, never recordings, and one switch in Settings stops it.
What people ask before installing
Where does recognition actually happen?
On Mac it runs in the cloud for the best accuracy: the audio goes to api.hearflow.it and from there to Groq (United States), and comes back as text. If that service cannot be reached and you have downloaded the local Whisper model, the app falls back to your Mac. On Windows it always runs on your PC; on iPhone Apple’s speech recognition does it, on the device when the language model is there.
What ends up written on my Mac?
Everything sits in your user library: the dictionary in the app’s preferences; the history of up to 90 days with the audio of your dictations, statistics and, if downloaded, the local models in ~/Library/Application Support/HearFlow; a log in ~/Library/Logs.
Is anything ever sold or handed over?
No. We do not sell data. What we keep — your account, usage counters and the content of the features you leave on (Audios sync, MCP, review) — is never sold, and AI apps read it only if you connect them. Subscriptions pay for the work.
Does my voice end up in a training set?
No. Your audio is not used for training. Transcription keeps no audio; synced audio is kept only to sync your devices, for up to 90 days. Sending dictations for review shares text (raw, final and your edits), never the recording, and Settings turns it off.
How do I confirm this on my own?
Dictate with Little Snitch or a packet sniffer open: on Mac dictation goes to api.hearflow.it. You will also see hearflow.pages.dev (updates and feedback), Hugging Face or the independent mirror hf-mirror.com (model downloads) and, while those features are on, our hearflow-sync and hearflow-improve services on workers.dev.
Who can see what I dictate?
The text is typed straight into the app you are using. On Mac it passes through our server to Groq for transcription and cleanup without being stored, and it is kept only by the features you leave on — Audios sync, sending for review, the MCP server. In our work we read only the dictations sent for review and the reports and feedback you send us.
Do you sign agreements for companies?
Write to info@hearflow.it: the processing is exactly the one described in the Privacy Policy, providers included, so the paperwork is short.
What stays behind if I uninstall?
Dragging the app to the Bin leaves ~/Library/Application Support/HearFlow (history, audio, models), the preferences (where the dictionary lives), the log in ~/Library/Logs and the Keychain token: delete them by hand. Disconnect your agents first, so the hooks and the codex launcher are removed from Claude Code and Codex. Copies on our servers stay until they expire: use Delete synced audios, or delete your account, to remove them at once.
Speak freely. You decide what is kept.
Hold the shortcut, talk, keep writing. The words land in the app you were already using; what we keep depends on the switches you leave on, each one described in the Privacy Policy.
7 days of Pro · then 2,000 words a week, with no expiry · no card needed