Apple Foundation Models and Image Playground

Implement reliable Apple Intelligence features with Apple's on-device text and image generation APIs across all supported platforms.

Model Limitations (CRITICAL — Read First)

The on-device model is ~3B parameters with a 4,096-token context window. It excels at summarization, extraction, classification, tagging, composition, and revision. It cannot reliably do:

• World knowledge / factual Q&A — will hallucinate
• Math and arithmetic — unreliable at multi-step calculations
• Code generation — cannot produce correct code
• Long-form writing (>200 words) — context window too small

Before writing any code, check references/model-capabilities-and-limits.md to confirm the task is within the model's capabilities. If it is not, design an escalation path (tool, backend model, or user disclosure).

Quick Reference