// Generated from the feature catalog
Screenshot & Diagram Analysis
Speaks and Listens Without the Cloud · Shipped · workstation
Paste, drag, or capture an image for the vision model to analyze and explain — Wayland-aware screen capture included.
Image analysis via the vision model tier — paste, drag-and-drop, or screen-capture an image into a turn for the agent to analyze. The screen capture module works on both X11 and Wayland compositors, and on macOS via ScreenCaptureKit.
Images are routed to the vision model tier. The agent can describe, annotate, and reason about screenshots, architecture diagrams, error dialogs, and configuration UIs. The vision tools support multi-image analysis and region-of-interest cropping.
Limits and invariants
Section titled “Limits and invariants”Images are processed locally by default — cloud vision endpoints require explicit configuration.
Where this lives
Section titled “Where this lives”halbert_core/halbert_core/tools/vision_tools.pyhalbert_core/halbert_core/vision/screen_capture.pyhalbert_core/halbert_core/dashboard/routes/vision.py