Practical overview
When it fits
Use it when developing with Apple’s Core AI framework and you need a workbench spanning recipe discovery, asset inspection, conversion, runtime experiments or benchmark evidence.
Who it is for
- Developers exploring Apple Core AI model assets and conversion recipes
- Apple-platform developers who want to run model functions and record benchmark evidence
- Teams organizing model artifacts, provenance, recipe revisions and run records in projects
Documented uses
- Browse pinned Apple model recipes and their export instructions.
- Convert a recipe on macOS using a local model-repository checkout.
- Inspect .aimodel metadata, functions, compute types and cache profiles.
- Run a model function, benchmark it and export evidence JSON.
- Package a Swift integration export or organize assets and provenance in a project.
Requirements & limits
The repository is public and MIT-licensed. The README documents Xcode 27 beta, Swift 6.4, iOS 27 and macOS 27; its quick start provides a macOS scheme and a scheme for a physical iPhone or iPad.
- Apple recipes are conversion instructions, not included model weights; the documented recipe workflow uses a local apple/coreai-models checkout.
- The app does not download models or bypass upstream licenses.
- Recipe Studio validates recipe and pipeline contracts but does not execute them.
- The README distinguishes Core AI from Core ML and FoundationModels; only the Qwen adapter bridges through FoundationModels.
- The README identifies accepted upstream access for SAM 3 and user-provided authentication for gated models.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Build the macOS app from the checked-in Xcode project
Open the project and build the macOS app using the documented Xcode 27 beta setup.
Before you start
- A local checkout of the repository
- Xcode 27 beta
- An arm64 macOS build destination
Steps
Set DEVELOPER_DIR to /Applications/Xcode-beta.app/Contents/Developer and run xcodebuild -version.
Open CoreAIFrameworkLab.xcodeproj.
Build the CoreAILabMac scheme for platform=macOS,arch=arm64, using the documented derived-data path ./build/Xcode27.
Open build/Xcode27/Build/Products/Debug/CoreAILab.app.
Expected outcome
The documented workflow produces a macOS app build at the specified path.
Guide limitations
- The README specifies Xcode 27 beta and an arm64 macOS destination.
- This guide covers building the app, not running a model.
Inspect and benchmark an imported .aimodel
Use the documented Function Workbench sequence to benchmark a model asset and export evidence.
Before you start
- The Core AI Framework Lab app
- An available .aimodel asset
Steps
Open the .aimodel in Asset Inspector.
Select an automatic, CPU-only, GPU-preferred or Neural-Engine-preferred specialization profile.
Open Runtime Studio → Function Workbench.
Generate bounded deterministic NDArray inputs and run a stateless function.
Run the benchmark and choose Export Evidence JSON.
Expected outcome
The workflow produces benchmark evidence as a JSON export.
Guide limitations
- This sequence requires an imported .aimodel.
- It describes a stateless function workflow; it does not establish results for every model or device.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- rudrankriyam/Core-AI-Framework-Lab · Official homepage title · retained in its source language