ahmadhassan-bted/b

Purpose and Entry Point

B is a Python/PyQt6 desktop companion that observes desktop context, accepts typed or spoken input, runs LLM reasoning, and renders speech, text, emotion, and movement in a transparent overlay. Startup is main.py, which composes modules, registers global hotkeys, and drives the 60 FPS application tick.

Event-Driven Core

core/bus.py is the architectural boundary between subsystems: modules publish and subscribe through a thread-safe Qt signal-based EventBus rather than calling one another directly. main.py owns process lifecycle and bus wiring. Core control flow is perception/input → bus → cognition/state → bus → visual/audio/motion outputs.

Perception and Context Acquisition

Active application changes originate from sensors/window_tracker.py. vision/semantic.py extracts structured UI context via UI Automation and quality/cooldown logic. When semantic extraction is unavailable, vision/mss_capture.py captures the screen and uses OCR as a fallback. These stages emit context events rather than exposing platform-specific APIs to the brain.

User Input and Work Modes

ui/input_box.py provides typed interaction; audio/ears.py provides microphone transcription using Faster-Whisper with voice activity detection. brain/work_mode.py defines the focused/proactive work-mode behavior activated by global hotkey. Input is normalized into bus events consumed by cognition, keeping UI and audio capture independent from reasoning.

Cognition, State, and Autonomy

brain/llm.py is the CognitiveEngine: it combines context/history, invokes either Groq cloud inference or local llama-cpp, and parses streamed output. brain/context.py manages contextual state, while brain/soul.py owns emotional/conversation state and sentence-ready stream buffering. brain/autonomy_loop.py schedules unsolicited thoughts only when timing and context quality permit. Cognition publishes speech, expression, and movement requests; it does not directly render or play them.

Expression and Overlay Runtime

ui/overlay.py hosts the transparent, click-through desktop presentation. ui/face.py renders the animated face, ui/chat.py displays spoken text, and physics/kinematics.py translates movement requests into eased/Bezier motion. These are Qt-driven, 60 FPS output consumers coordinated through events and state transitions.

Audio and External Dependencies

audio/speaker.py turns generated sentences into Piper ONNX speech and applies Pedalboard DSP effects; packaged voice assets are voices/b_voice.onnx and voices/b_voice.onnx.json. External services/runtime boundaries include Groq API credentials from .env, optional local GGUF/llama-cpp inference, Windows UI Automation/Win32 hooks, Tesseract OCR, MSS screen capture, Faster-Whisper, and PyQt6. Primary target is Windows 10/11, with experimental Linux/macOS support.