Skip to content
The Atlas

Voice input

Press ctrl+r in the chat, speak, press Enter. The recognized text is inserted into the input box (it is not sent), so you can edit it first.

Setup

Section titled “Setup”

  1. Install ffmpeg (it records the microphone):

    Terminal window
    winget install Gyan.FFmpeg

    Open a new terminal afterwards. If ffmpeg is installed somewhere The Atlas cannot find, set ATLAS_FFMPEG to its full path.

  2. Connect Groq for speech recognition (free):

    Terminal window
    atlas connect groq

    The Atlas uses Groq’s whisper-large-v3-turbo. Without a Groq key it falls back to OpenAI whisper-1 if an OpenAI key is connected.

In the chat

Section titled “In the chat”
Key Action
ctrl+r or /voice Start recording
Enter Stop and transcribe
Esc Cancel; nothing is uploaded

Text you had already typed is kept; the recognized text is appended.

From the shell

Section titled “From the shell”

atlas voice records once and prints the text to stdout, so you can pipe it:

Terminal window
atlas voice
atlas voice -l ru                 # language hint: uz, ru or en
atlas voice -s 30                 # stop after 30 seconds (default 120)
atlas voice | atlas ask           # ask by voice

Choosing a microphone

Section titled “Choosing a microphone”
Terminal window
atlas voice devices               # list microphones; the one in use is marked
atlas voice device 2              # remember microphone number 2
atlas voice device "USB Mic"      # or by name
atlas voice device auto           # back to the system default
atlas voice -d 2                  # use one just for this recording

The choice is saved in ~/.the-atlas/voice.json.