Map agent intent to desktop coordinates.
Compare normalized vision bounds (0–1000) against Cocoa logical points, Wayland fractional surfaces, and OS accessibility trees without dispatch surprises.
Transform normalized inputs to hardware pixels.
Agent vision models emit point coordinates in normalized units [0..1000]. Execute the multi-stage pipeline below to calculate precise OS dispatch targets.
Hardware Coordinate Mapping
Interactive Target Pad
Norm(630, 380)Synthesize intent across platform layers.
A coordinate without OS semantic backing leads to failed clicks on detached surfaces, popup layers, or unscaled window viewports.
Transform pipeline verification
On macOS, Quartz window server coordinates operate in logical points where top-left is (0,0). High-DPI Retina displays apply a backing store scale factor (typically 2.0x). Synthetic input via CGEventCreateMouseEvent uses logical points, while accessibility queries via AXUIElementCopyAttributeValue report element frame origins in screen coordinates.
Wayland intentionally restricts global coordinate discovery and synthetic event injection across arbitrary client surfaces. Automation agents must interface with Compositor protocols (such as wlr-virtual-pointer or XDG Desktop Portals) alongside AT-SPI D-Bus trees to discover fractional scale surfaces.
Under X11, global coordinates are directly addressable via the root window, and synthetic events are dispatched through XTestFakeMotionEvent. However, multi-monitor xrandr configurations with mixed DPI scaling require manual coordinate offsetting to prevent pointer clipping.
Export mapping specifications for agent pipelines.
Download current transform parameters as a structured plain-text specification ready for integration with OS agent test runners.