Full Computer Vision Suite

Knows What’s Wrong Before You Do · Shipped · workstation, home

Screen capture (Wayland-aware), ambient webcam monitoring, OCR text extraction, and motion detection.

A comprehensive local vision subsystem that extends far beyond simple image pasting. Screen capture works on both X11 and Wayland compositors. Ambient webcam monitoring detects environmental changes. OCR extracts text from screenshots and documents. Motion detection identifies physical movement in camera feeds.

Each vision module operates independently with graceful degradation — if onnxruntime is not installed, the pipeline starts but produces no observations (no crash). Screen capture uses platform-specific APIs (Wayland portal, X11 capture, macOS screencapturekit). OCR runs locally via Tesseract or ONNX models.

Limits and invariants

All vision processing is strictly local — no images are ever transmitted to cloud services unless the user explicitly configures a cloud vision model.

Where this lives

• halbert_core/halbert_core/vision/screen_capture.py
• halbert_core/halbert_core/vision/ambient_webcam.py
• halbert_core/halbert_core/vision/ocr.py
• halbert_core/halbert_core/vision/motion.py
• halbert_core/halbert_core/tools/vision_tools.py