Voice input
Dictate into chats and notes, with your words appearing as you speak.
Tap the microphone and speak. Your words appear as you talk, about a second behind your voice. When you pause, the sentence is checked once more and its final text replaces the live words.
How to use it
- Tap the microphone in Your log or an AI agent chat and allow microphone access.
- Speak naturally. Recording continues while each sentence is transcribed.
- While you talk, the microphone glows with your voice. A thin spinning ring around it means some of your words are still on the way.
- Tap the mic again to stop and process the last section.
- Read and edit the result, then send it yourself.
Voice input never sends a message automatically. You can keep typing between sections, and new text appends to your draft.
In the note editor
On a computer, type / and choose Dictate. Use Stop or press Esc to finish. On a phone, use the microphone in the editor toolbar. Each completed section inserts at the cursor and saves through the normal editor. Individual property fields do not have microphone buttons.
Language
Voice input understands 90+ languages. It is most accurate in English, Russian, Polish, Latvian, Estonian, Ukrainian and the main Western European languages; Lithuanian is also highly accurate. Auto-detection is the default. To give transcription a language hint, open Settings → Profile → Voice input language, choose your language, and save. Leave auto-detection on when switching between languages. Names, numbers and mixed-language speech can need corrections, so check the result before sending.
Processing and retry
Recording keeps going until you tap the microphone again. If a sentence fails to transcribe, it is retried by itself while you keep talking. If it still fails, use Retry (or tap the microphone again). Already completed text stays in place. Keep this screen open until pending audio finishes: leaving or cancelling discards audio that has not been uploaded.
A timed-out request may already have used provider credits. Retrying an unfinished section can make another paid request. Completed, saved sections are reused.
Recorded uploads support up to 25 MB and two hours of processing. Longer recordings are processed in sections. If the service reports incomplete transcription, retry the same file to resume; incomplete text is not pasted as the whole recording.
Privacy and payment
Audio goes through OpenRouter to ElevenLabs Scribe. Dictated sentences are not stored: only the text lands in your draft. Uploaded recordings and their raw transcription responses are stored privately so unfinished work can resume. Eligible BYOK uses your own OpenRouter key; managed AI uses your Kaizendex allowance. If your account or credential cannot be verified, paid transcription stops.
Troubleshooting
- Microphone denied: enable microphone permission in your browser or device settings.
- No text yet: pause briefly, or stop to send the remaining audio. Silence produces no text.
- Transcription paused: retry before leaving this screen.
- Update required: install the current app; older live-dictation clients must upgrade.
- No microphone button: the browser may not support audio capture. Typing remains available.