Where does my voice go when I dictate with Zefmi?

1. You hold the key and speak

The app records while you hold the shortcut (Alt + Space by default) or while hands-free mode is on. Nothing is recorded otherwise.

2. Your audio goes to a speech service

The app streams the audio to our relay, which passes it to AssemblyAI. Users in the EU are routed to AssemblyAI's EU endpoint. If AssemblyAI is down, the relay can use Deepgram as a backup. The text comes back to the app. Our relay stores no audio and no text, and it logs status codes only.

3. The text is cleaned up, if you want

With AI clean-up on, the text is sent through OpenRouter to Google Gemini 2.5 Flash Lite, with OpenAI GPT-4.1 nano as the backup. We set OpenRouter's data collection option to deny, so only providers that do not log or train on prompts are used. In Word for word mode this step is skipped.

4. The text is typed at your cursor

Zefmi types by pasting, the way Ctrl+V does. It marks its clipboard write so Windows keeps it out of clipboard history and the cloud clipboard, then puts your previous clipboard back.

Does Zefmi store my audio?

Not on our relay. On your own PC, the app keeps recent recordings for 7 days by default so you can replay them. If the live stream fails, a one-shot upload path sends the recording to AssemblyAI and deletes the transcript and the audio there once it has the text.

Does Zefmi read my screen or my clipboard?

It can read the text around your cursor as context, and you can switch that off in settings. It never reads password fields. It does not take screenshots and does not log your keys beyond the shortcut. The clipboard is used only for the paste described above.

Does Zefmi train AI on my words?

No. How long each speech or AI service keeps data depends on its own settings and policy, so we link each one in the table below and do not restate it in our own words.