Core AI Debugger
A standalone Mac app to visualize, run and validate Core AI models across Apple platforms.
Free
Overview
Core AI Debugger is a standalone macOS app for working directly with Core AI model files (.aimodel). It shows a model's operations as a connected graph, and when the model was exported from PyTorch with debug metadata, it maps each operation back to its PyTorch module and Python source line.
You can specialize and run a model on your Mac or a paired device to inspect the output tensor of each operation, then compare the results against a PyTorch reference run or a second configuration. The debugger finds sync points, scores their similarity with metrics such as PSNR, and shows where the outputs diverge. It works alongside the Core AI debug gauge in Xcode and the Core AI template in Instruments.
Model graph
Shows a model's operations as a graph, mapped back to the PyTorch module and source line.
On-device runs
Specializes and runs a model on a Mac or paired device, then shows each tensor output.
Reference comparison
Finds sync points against a PyTorch reference run and scores similarity, such as PSNR.
Xcode handoff
Takes events captured by the Core AI debug gauge in Xcode for closer inspection.
History
How it grew up.
Introduced
Core AI Debugger is introduced at WWDC26 as a new standalone app for Core AI models.
Beta download
Apple offers the debugger as a beta download for macOS 27 from the Apple Developer website.