Nara Mobile AI Lab is a developer-focused test bench for exploring Apple built-in AI capabilities directly on iPhone.
The app helps developers, builders, and technical teams understand how on-device AI performs in real app workflows before investing in a larger implementation.
For the most accurate test, turn on Airplane Mode before running experiments. This helps verify whether results are coming from the device’s built-in AI capability instead of a cloud service.
With Nara Mobile AI Lab, you can test practical AI workflows such as:
Speech-to-text with structured JSON extraction
Vision and OCR extraction from photos
Prompt response testing
Mock tool and callback execution
User-defined tool function testing
LocalKnowledge search testing
Object detection using preset Core ML models
Image classification using preset Core ML models
Prompt to Image using Image Playground
Latency, validation, manual scoring, and notes
Exporting results for review or project planning
The app includes built-in templates and one-time page hints so you can quickly test common developer scenarios without building a full prototype first. You can inspect raw extraction, AI output, validation status, benchmark timing, manual score, and notes.
Nara Mobile AI Lab is especially useful for evaluating whether on-device AI is suitable for privacy-focused apps, offline-first workflows, mobile automation, field tools, form extraction, image analysis, local knowledge search, and AI-assisted business software.
Core ML Vision Tests
Nara Mobile AI Lab includes preset Core ML model tests for object detection and image classification. The current preset models include YOLOv3TinyFP16 for object detection and FastViTMA36F16 for image classification. These tests provide practical developer examples for reviewing on-device model behavior, latency, and output format.
Tool Calling Tests
The app includes mock tool-calling tests so developers can evaluate whether the Apple Foundation Model chooses the correct function, passes reasonable arguments, and returns a useful result. Write-style actions are mocked for safety, allowing the app to demonstrate tool-calling behavior without performing real external changes.
LocalKnowledge Tests
The app includes local knowledge search testing so developers can explore how user-defined local information can be used in offline-first AI workflows. Test data stays on the device unless the user manually exports it.
Nara Mobile AI Lab is a testing and learning tool. AI output may be incomplete or inaccurate and should always be reviewed by the user.
