B is a PyQt6 desktop overlay with an animated face. It reads the active window through Windows UI Automation, with Tesseract OCR as a fallback. It sends that context to Groq's cloud API or a local llama-cpp model, and replies by voice (Piper TTS) and chat bubble. Speech input uses Faster-Whisper, and a work mode lets it offer suggestions unprompted.
Who it is for: Windows users wanting a voiced, screen-aware desktop assistant
Strengths
Supports both Groq cloud inference and local llama-cpp models
Screen context via UIA tree walking, with OCR fallback for incompatible apps
Voice loop: Faster-Whisper input with VAD, Piper TTS output
Global kill switch registered at Win32 level for instant exit
Weaknesses
Windows 10/11 is the primary target; Linux and macOS are experimental
UIA screen reading is Windows-specific, so context is weaker elsewhere
No window close button; the hotkey is the only way to exit
No tagged releases; README lists status as active development
What it needs
RAM ≥ 16 GB
GPU optional
Needs Groq API key, Tesseract OCR, Piper ONNX voices, Faster-Whisper