API ReferenceLocal APIs

useLocalModel

React hook — on-device LLM chat over WebGPU.

Auto-generated from JSDoc in packages/react/src/local/useLocalModel.ts. Do not edit — run npm run docs:api.

useLocalModel

function

function useLocalModel(modelId: string, opts: UseLocalModelOptions): UseLocalModel

MODELS

variable

const MODELS: BrowserModel[]

LocalMessage

interface

interface LocalMessage

`role: user | assistant`

`content: string`

UseLocalModel

interface

interface UseLocalModel

`status: LocalModelStatus`

`progress: number`

Weight-download progress, 0–1 (only meaningful while status === "loading").

`messages: LocalMessage[]`

`error?: Error`

`load: () => Promise<LanguageModel>`

Begin downloading + warming the model. Idempotent; safe to call eagerly for a progress bar.

`send: (text: string) => Promise<void>`

Send a user turn and stream the assistant reply into messages.

`stop: () => void`

Abort the in-flight reply.

`reset: () => void`

Clear the conversation.

UseLocalModelOptions

interface

interface UseLocalModelOptions

`accessToken?: string`

Hugging Face token for gated/private repos.

`autoLoad?: boolean`

Start downloading the model on mount instead of waiting for the first send/load().

LocalModelStatus

type alias

type LocalModelStatus = idle | loading | ready | generating | error

AI SDK provider

The valora() AI SDK provider — on-device LLM inference, no server.

useLocalTTS

React hook — on-device text-to-speech (Kokoro) over WebGPU.

On this page

useLocalModel MODELS LocalMessage role: user | assistantcontent: stringUseLocalModel status: LocalModelStatusprogress: numbermessages: LocalMessage[]error?: Errorload: () => Promise<LanguageModel>send: (text: string) => Promise<void>stop: () => voidreset: () => voidUseLocalModelOptions accessToken?: stringautoLoad?: booleanLocalModelStatus

Valora is local-first

No API key, no server — everything in this doc runs on-device.