useGenAITasks

Source: packages/sdk/src/ai/useGenAITasks.ts

Dedicated on-device GenAI task clients on ML Kit and AICore, separate from the chat surface: summarize, proofread, rewrite and describe an image. Every call is measured and reported with hardware provenance. Unlike useGeminiNano’s equivalents, these resolve to null on failure instead of throwing, and keep the last result in state. Each task takes its own text or image input; nothing is read on mount.

Signature

function useGenAITasks(): {
 isRunning: boolean;
 error: string | null;
 source: TelemetrySource;
 summaryResult: SummarizeResult | null;
 proofreadResult: ProofreadResult | null;
 rewriteResult: RewriteResult | null;
 imageDescriptionResult: ImageDescriptionResult | null;
 summarize: (text: string, options?: SummarizeOptions) => Promise<SummarizeResult | null>;
 proofread: (text: string) => Promise<ProofreadResult | null>;
 rewrite: (text: string, tone?: TaskTone) => Promise<RewriteResult | null>;
 describeImage: (imageInput: string, style?: 'detailed' | 'caption' | 'labels' | 'concise') => Promise<ImageDescriptionResult | null>;
};

Outputs

Field	Type	Description
isRunning	boolean	true while any of the four tasks is in flight. They share one flag, so run them one at a time.
error	string | null	Why the last task failed, including “PixelNano module is unavailable in this environment” on web.
source	TelemetrySource	'hardware' when the PixelNano module is present, 'unavailable' otherwise.
summaryResult	SummarizeResult | null	Last summary: { summary, latencyMs, engine, source }.
proofreadResult	ProofreadResult | null	Last proofread: { correctedText, suggestions, latencyMs, engine, source }.
rewriteResult	RewriteResult | null	Last rewrite: { rewrittenText, suggestions, latencyMs, engine, source }.
imageDescriptionResult	ImageDescriptionResult | null	Last description: { description, finishReason, latencyMs, engine, source }.

Functions

Function	Inputs	Returns	Description
summarize(text, options?)	text: string — article or conversation transcript. options.inputType?: 'article' | 'conversation' — how to read the input. options.outputType?: 'one_bullet' | 'two_bullets' | 'three_bullets' — how long the summary should be.	Promise<SummarizeResult | null> — null on failure, with error set	Runs on-device; no text leaves the phone.
proofread(text)	text: string — the text to correct	Promise<ProofreadResult | null>	Returns corrected text plus the individual suggestions.
rewrite(text, tone?)	text: string. tone?: TaskTone — 'elaborate' | 'emojify' | 'shorten' | 'friendly' | 'professional' | 'rephrase', default 'professional'.	Promise<RewriteResult | null>	Transforms tone or length while keeping the meaning.
describeImage(imageInput, style?)	imageInput: string — a file URI or base64 image. style?: 'detailed' | 'caption' | 'labels' | 'concise', default 'concise'.	Promise<ImageDescriptionResult | null>	On-device multimodal description, for alt text or a caption.