useEmbeddings

Executes on-device text embedding generation on the Google Tensor EdgeTPU / NPU through ML Kit GenAI. Produces normalized 512-dimensional floating point vectors from input text in milliseconds. Includes an in-memory cosine similarity calculation helper to compare semantic proximity between vectors entirely offline.

Signature

useEmbeddings(): EmbeddingsTelemetry

Outputs

Field	Type	Description
`isAvailable`	`boolean`	Whether the on-device text embedding model is installed and ready for inference.
`isLoading`	`boolean`	Whether an embedding inference operation is actively computing on the NPU/TPU.
`vectorDimension`	`number`	Dimension size of the output embedding vector (512).
`error`	`string \| null`	Error message if embedding inference failed.
`source`	`TelemetrySource`	Data provenance: ‘hardware’ or ‘unavailable’.
`embed`	`(text: string) => Promise<number[]>`	Computes a normalized float vector embedding for the provided text.
`cosineSimilarity`	`(vecA: number[], vecB: number[]) => number`	Computes cosine similarity between two float vectors (-1.0 to 1.0).

Functions

Function	Inputs	Returns	Description
`embed(text)`	none	`void`	Generates a 512-dimensional embedding vector from input text.
`cosineSimilarity(vecA, vecB)`	none	`void`	Scores cosine similarity between two embedding vectors.

Example

import { useEmbeddings } from '@pixelkit-labs/sdk/mlkit';

function SimilarityDemo() {
  const { isAvailable, embed, cosineSimilarity } = useEmbeddings();
  const compare = async () => {
    const vecA = await embed('Google Pixel 11 Pro');
    const vecB = await embed('Android Smartphone');
    const score = cosineSimilarity(vecA, vecB);
    console.log('Similarity score:', score);
  };
  return <Button title="Compare Texts" onPress={compare} />;
}