Use Codmir as a local voice assistant to capture the screen, record audio, transcribe speech, and control your desktop.

Voice Desktop Assistant (Local)

Codmir can act as a local voice-driven desktop assistant. Say the wake word “codmir” to start a privacy-first session that:

  • Captures low-FPS screenshots with change detection
  • Shows blue corner lines to indicate visibility
  • Records microphone audio and transcribes speech offline
  • Optionally executes desktop intents (open URLs/apps, type, press keys)
  • Saves a chronological session bundle (audio + frames + transcript)

Architecture Overview

  • Wake word → Orchestrator → Screen Monitor + Audio Recorder → Transcriber → Desktop Automation → Session Persistence
  • All components run locally via Docker.

Quick Start

  • Compose file: docker/compose.codmir.yml
  • One-time X11 grant (X11 only): 
    xhost +local:root
docker compose -f docker/compose.codmir.yml up --build -d
  • Say “codmir” to trigger a session. View sessions: 
    curl -sS http://localhost:8096/sessions | jq

Wayland vs X11

  • Overlay and desktop control require X11/XWayland.
  • On pure Wayland, disable overlay and use limited control; Wayland backends are planned.

Learn More