Skip to main content

Record audio

Audio is useful when speaking is faster than typing or when tone and detail matter. A recording can stand alone or live inside a task.

Capture a recording

  1. Open the journal or the task that should own the recording.
  2. Use the + action and choose Audio Recording.
  3. Review the available processing options.
  4. Start recording, speak clearly, then stop when finished.
All screenshots
ScreenshotMobile view
Active audio recording sheet showing the live visualizer, elapsed time, pause, discard, stop, and speech recognition controls

During capture, the sheet keeps elapsed time and the live input visualizer in view. You can pause and resume without ending the recording, discard the take, or choose Stop to save it. When task-aware transcription is available, the Speech Recognition option controls whether processing should begin after the recording is saved.

When an inference profile and transcription skill are configured, Lotti can transcribe the recording after capture. Task recordings may also offer task-aware processing such as checklist or summary updates.

Dictate useful task context

For task-aware processing, say the outcome and then name concrete actions, people, dates, or constraints. “Confirm the Friday cargo window, then ask Mission Control about the launch approval” gives Lotti far more useful context than “sort out the penguin thing.” A short pause between distinct actions also makes the resulting transcript easier to review.

Task-aware processing can turn that recording into a transcript, a refreshed report, or checklist suggestions. These are reviewable AI outputs, not a claim that the task changed by itself: check names, dates, and proposed changes before you accept them. For difficult audio, move somewhere quieter, keep the microphone reasonably close, and speak at a natural, steady pace.

Review saved transcripts

Open an audio entry, use its menu, and choose Speech Recognition. The review sheet keeps every saved transcript with the recording instead of replacing an earlier result. Each row shows when processing ran, how long it took, the detected language, and the provider and model that produced it.

All screenshots
The source recording remains visible behind the sheet, while each recognition result keeps its own provenance.
ScreenshotMobile view
Speech Recognition sheet over a Project Waddle audio entry with English selected and two saved transcript results from Mistral and Whisper

The language selector currently offers Auto, English, and Deutsch. Use Auto when the provider should detect the language, or select a language before running recognition when you already know it. The choice is stored on the audio entry.

Select the double-chevron on a result to reveal the complete, selectable text. Expanding a row also reveals its delete action. Deleting a transcript removes that recognition result, not the source recording or the other transcripts.

All screenshots
Review the full wording and provider provenance before using generated text as task context.
ScreenshotMobile view
Expanded Project Waddle transcript describing Europa sardine cargo, 37 emperor penguins, and the zero-gravity fish feeder

Choose automation deliberately

Processing options are requests to configured AI skills. The selected provider may be local or remote, so verify the active inference profile before using sensitive material. Review generated text and proposed task changes rather than treating them as source truth.

See AI provider setup to control where inference runs.