SystemLanguageModel

Represents the on-device Foundation Models language model. Provides availability checking and model configuration.

Constructor

new SystemLanguageModel(options?: {
  useCase?: SystemLanguageModelUseCase;
  guardrails?: SystemLanguageModelGuardrails;
})

Parameter	Default	Description
`useCase`	`GENERAL`	Model use case
`guardrails`	`DEFAULT`	Guardrail configuration

Methods

`isAvailable()`

Synchronously checks if the model is ready.

isAvailable(): AvailabilityResult

Returns { available: true } or { available: false, reason: SystemLanguageModelUnavailableReason }.

`waitUntilAvailable()`

Polls until the model is available or the timeout expires.

waitUntilAvailable(timeoutMs?: number): Promise<AvailabilityResult>

Parameter	Default	Description
`timeoutMs`	`30000`	Maximum wait time in milliseconds

`supportsLocale()`

Check whether the model supports a given locale.

supportsLocale(localeIdentifier: string): boolean

model.supportsLocale("en_US"); // true
model.supportsLocale("ja_JP"); // true or false depending on model

`dispose()`

Releases the native model reference.

dispose(): void

Properties

`supportedLanguages`

Returns the locale identifiers the model supports (e.g. ["en-US", "es-ES"]).

readonly supportedLanguages: string[]

`contextSize`

The maximum number of tokens the model's context window can hold. All input — instructions, prompts, tool definitions, and responses — counts against this limit.

readonly contextSize: number

`tokenCount()` macOS 26.4+

Counts the tokens an input consumes against the context window. Asynchronous, and requires a macOS 26.4+ runtime.

tokenCount(input: TokenCountInput): Promise<number>

type TokenCountInput =
  | { prompt: string | PromptInput }
  | { instructions: string }
  | { tools: Tool[] }
  | { schema: GenerationSchema }
  | { transcript: Transcript };

Exactly one field applies per call — the C bridge exposes a separate entry point for each kind of input:

await model.tokenCount({ prompt: "Summarize this article." });
await model.tokenCount({ instructions: "You are a helpful assistant." });
await model.tokenCount({ tools: [weatherTool] });
await model.tokenCount({ schema: ContactCard.schema });
await model.tokenCount({ transcript: session.transcript });

Useful for staying inside contextSize before sending a request — tool definitions and schemas are often larger than they look. Measured against the on-device model, a five-word prompt costs 15 tokens while a single-argument tool definition costs 83.

Enums