Cross-Platform Optimization & AI Engine Upgrade (Gemini 3.1 Flash Live)

[archpulse]

Summary
Major refactor to ensure 100% compatibility across Linux (Wayland/X11), Windows, and macOS. Fixed critical crashes,
upgraded the AI model, and implemented advanced vision logic.

Key Changes

1. Core & Stability

• Thread-Safe UI: Implemented a queue.Queue architecture for Tkinter to prevent Segmentation Faults on Linux.
• AI Engine Upgrade: Switched to gemini-3.1-flash-live-preview and upgraded google-genai to v1.73.1.
• Reliability: Added exponential backoff for reconnections and fixed the 1007 media_chunks API error.

2. Enhanced Vision (Wayland Compatible)

• Video Sequence: Jarvis now captures a 2-frame sequence (downscaled to 720p) for better analysis, matching Gemini
  3.1's preferred input.
• Linux Vision: Integrated xdg-desktop-portal and dbus support for seamless screen capture on Wayland.

3. Platform-Agnostic Actions

• System Controls: Added support for PipeWire/PulseAudio/ALSA on Linux; improved brightness and lock-screen logic for
  all OS.
• Smart App Launching: Added URL support (e.g., "Open YouTube") and deep search for Linux binaries via .desktop
  files.
• New Tools:
  • get_system_status: Real-time CPU/RAM monitoring.
  • media_control: System-wide play/pause and track skipping.

4. Optimization

• Refactored Tool Registry: Replaced massive if/elif blocks with a dynamic TOOL_MAP for faster execution.
• Audio Robustness: Increased buffer sizes and added safe frame-dropping to prevent QueueFull errors.

Verification

• Tested on Linux (Wayland) using uv run main.py.
• Verified voice/text responses and screen analysis accuracy.
• Confirmed all temporary debug artifacts were removed.