Docs › Extending AstraCode
Dictation
Talk to the agent instead of typing: pick a microphone, record, and get text in the composer.
Press the microphone button under the composer to record, and press it again to stop. What you said arrives as text in the prompt box, where you can edit it before sending.
Choosing a microphone
The chevron beside the button lists every input your system offers: built-in, USB, headset, and on macOS your iPhone when Continuity is available. Outputs are never listed as inputs.
Device names only become visible once you have granted permission; before that the system reports them unnamed, so the list shows what is there and names what it can.
Permission
The first recording raises your operating system's microphone prompt against AstraCode. A refusal there is final and has to be undone in your system settings. AstraCode cannot re-ask.
AstraCode only listens while you are recording. Opening the device list does not start a recording; the stream it briefly opens to read device names is stopped immediately.
The language you speak
By default the model works out the language from the audio. That is right for most people and least reliable for the languages with the least training data, and for short clips. Nepali is often transcribed as Hindi or English for exactly that reason.
Set astracode.dictation.language to the language you dictate in and the guess disappears. It is a separate question from the language AstraCode replies in, which follows whatever you write.
What happens to the audio
The recording is sent to the AstraCode gateway, transcribed, and the text comes back. Nothing is stored after transcription. Recordings are capped at a minute: dictation is for a prompt, not a meeting.
All documentation
Get started
Agent
- Agent overview
- Approvals and permissions
- Plan mode
- What the agent can do
- Running commands
- Browser tools
- Checkpoints and undo
- Infrastructure as code
- Running several agents at once