Availability, model info & diagnostics

`getAvailability(model?)`

const availability = await AppleFoundationModels.getAvailability();              // system model
await AppleFoundationModels.getAvailability('privateCloudCompute');             // iOS 27
await AppleFoundationModels.getAvailability('auto');                            // which model `auto` would use

Never throws. Returns:

{ status: 'available', available: true, model: 'system' }
{ status: 'unavailable', available: false, reason: 'appleIntelligenceNotEnabled', nativeReason: '...' }

`reason`	Apple source	What to do
`deviceNotEligible`	`SystemLanguageModel.Availability.UnavailableReason.deviceNotEligible` (also PCC)	Hide the feature permanently.
`appleIntelligenceNotEnabled`	`.appleIntelligenceNotEnabled`	Ask the user to enable Apple Intelligence.
`modelNotReady`	`.modelNotReady`	Assets downloading; check again later.
`systemNotReady`	`PrivateCloudComputeLanguageModel...UnavailableReason.systemNotReady`	PCC not ready; retry later.
`unsupported`	(Fourlyn)	Not iOS, iOS < 26, or no native module (Expo Go).
`unknown`	(Fourlyn)	A reason added by a newer OS; treat as unavailable.

Availability can change while the app runs (the user can disable Apple Intelligence, assets can be deleted). Re-check when the app returns to the foreground (useAvailability().refresh).

Native mapping: SystemLanguageModel.availability, PrivateCloudComputeLanguageModel.availability.

Model handles

const model = AppleFoundationModels.systemModel({ useCase: 'contentTagging', guardrails: 'default' });
await model.isAvailable();
await model.getContextSize();          // e.g. 4096
await model.getSupportedLanguages();   // ['de', 'en', 'es', ...]
await model.supportsLocale('fr-FR');
await model.tokenCount('Hello');       // iOS 26.4+
const session = await model.createSession({ instructions: '...' });

useCase maps to SystemLanguageModel.UseCase (general, contentTagging). guardrails maps to SystemLanguageModel.Guardrails (default, permissiveContentTransformations: use the latter only for transforming user-provided text, e.g. summarizing content that itself discusses sensitive topics).

`getModelInfo(model?)`

{
  model: 'system',
  availability: { status: 'available', available: true },
  isAvailable: true,
  contextSize: 4096,
  supportedLanguages: ['de', 'en', ...],
  variant: { id: 'core3', displayName: '...' } | null,   // iOS 27: 'core3' | 'coreAdvanced3'
  capabilities: { vision, guidedGeneration, reasoning, toolCalling } | null,  // iOS 27
}

Apple updates the model with OS releases (variant). Outputs can change between iOS versions: test prompts on each new OS and never assert exact model output.

`getDiagnostics()`

One call with everything relevant for a bug report: OS version, device model identifier, which SDK the app was compiled with, system/PCC model info, and a features map (tokenCounting, usage, images, dynamicProfiles, privateCloudCompute, visionTools, …). Use features to branch UI by capability instead of parsing OS versions.