Voice Mode

Voice mode lets you speak tasks instead of typing them.

Prerequisites

Install the voice extras:

pip install nimbus-ai[voice]

This installs:

sounddevice — microphone capture
numpy — audio processing
scipy — signal utilities
openai-whisper — local speech transcription (base model, ~140MB, downloaded on first use)

No audio is sent to external servers. Whisper runs entirely locally.

nimbus --voice

Or enable voice mode inside the REPL with the --voice flag on startup.

When you press Enter at the prompt, Nimbus starts recording:

  ●  recording... speak your task (press Enter to stop)

Recording stops when you press Enter again, or after 15 seconds of silence.

Nimbus transcribes your audio using the Whisper base model and displays the result:

  heard: add input validation to the user registration endpoint

If the transcription looks correct, Nimbus proceeds with the task. You can edit the transcribed text before approving.

During task execution, Nimbus listens for spoken approval keywords:

Spoken word	Action
"yes" / "approve"	Approve the plan or diff
"reject" / "cancel"	Reject

Requires microphone access. On macOS, grant permission when prompted by the system.
The base model is optimized for English. Other languages have reduced accuracy.
Voice mode does not work in SSH sessions without audio forwarding.