EmbeddingModel

Kotlin |Java

interface EmbeddingModel

Provides an interface for generating embeddings for text and multimodal Content.

Obtain instances via EmbeddingGeneration.getEmbeddingModels, and call close when the model is no longer needed to release underlying resources.

Typical usage:

val models = EmbeddingGeneration.getEmbeddingModels()
val model = models.firstOrNull() ?: return
models.drop(1).forEach { it.close() }
try {
  if (model.checkStatus() == FeatureStatus.DOWNLOADABLE) {
    model.download().collect()
  }
  val request = EmbeddingRequest(
    Content(listOf(TextPart("What is machine learning?"))),
    TaskConfig.RetrievalQuery(),
  )
  val result = model.generateEmbedding(request)
  val vector = result.embedding.vector
} finally {
  model.close()
}

Summary

Public functions
`suspend @FeatureStatus Int`	`checkStatus()` Checks the current availability status of the embedding feature.
`Unit`	`close()` Releases resources associated with the underlying embedding engine.
`Flow<DownloadStatus>`	`download()` Downloads the required model assets for the embedding feature if they are not already available.
`suspend EmbeddingResult`	`generateEmbedding(request: EmbeddingRequest)` Performs asynchronous content embedding on the provided input request.
`suspend EmbeddingMetadata?`	`getMetadata()` Returns the `EmbeddingMetadata` of the embedding model, or `null` if metadata retrieval fails or is unavailable.
`suspend String`	`getModelSignature()` Returns the model signature uniquely identifying the underlying embedding model and version.
`suspend Int`	`getTokenLimit()` Returns the maximum number of input tokens supported by this model in a single request.
`suspend Boolean`	`isDeprecated()` Returns `true` if the model is currently deprecated.
`suspend Unit`	`warmup()` Warms up the inference engine for use by loading necessary models and initializing runtime components, reducing latency for the first `generateEmbedding` call.

Public functions

checkStatus

suspend fun checkStatus(): @FeatureStatus Int

Checks the current availability status of the embedding feature.

Returns
`@FeatureStatus Int`	a `FeatureStatus` constant indicating the feature's readiness.

close

fun close(): Unit

Releases resources associated with the underlying embedding engine.

Callers should invoke this method when the model is no longer needed, including on any unused EmbeddingModel instances returned by EmbeddingGeneration.getEmbeddingModels.

download

fun download(): Flow<DownloadStatus>

Downloads the required model assets for the embedding feature if they are not already available.

The download starts when the returned cold Flow is collected. If the download fails, the flow emits DownloadStatus.DownloadFailed and terminates with the corresponding com.google.mlkit.genai.common.GenAiException.

Returns
`Flow<DownloadStatus>`	a cold `Flow` that emits `DownloadStatus` updates for download progress.

generateEmbedding

suspend fun generateEmbedding(request: EmbeddingRequest): EmbeddingResult

Performs asynchronous content embedding on the provided input request.

Parameters
`request: EmbeddingRequest`	the `EmbeddingRequest` containing input `Content` and optional `TaskConfig`.

Returns
`EmbeddingResult`	an `EmbeddingResult` containing the generated embedding vector and model signature.

Throws
`com.google.mlkit.genai.common.GenAiException`	if embedding generation fails or the request content is invalid.

getMetadata

suspend fun getMetadata(): EmbeddingMetadata?

Returns the EmbeddingMetadata of the embedding model, or null if metadata retrieval fails or is unavailable.

getModelSignature

suspend fun getModelSignature(): String

Returns the model signature uniquely identifying the underlying embedding model and version.

Embeddings generated by models with different signatures are incompatible and should not be directly compared. Callers that persist embeddings should store this signature alongside the vectors to detect model updates and re-index when the signature changes.

Returns
`String`	the model signature string.

Throws
`com.google.mlkit.genai.common.GenAiException`	if retrieving the model signature fails.

getTokenLimit

suspend fun getTokenLimit(): Int

Returns the maximum number of input tokens supported by this model in a single request.

Returns
`Int`	the maximum input token capacity, `-1` if the token limit is unknown, or `Int.MAX_VALUE` if there is no token limit restriction.

Throws
`com.google.mlkit.genai.common.GenAiException`	if retrieving the token limit fails.

isDeprecated

suspend fun isDeprecated(): Boolean

Returns true if the model is currently deprecated.

A model is considered deprecated if the current UTC date is on or after EmbeddingMetadata.deprecationDate. Returns false if model metadata or the deprecation date is unavailable.

warmup

suspend fun warmup(): Unit

Warms up the inference engine for use by loading necessary models and initializing runtime components, reducing latency for the first generateEmbedding call.

Throws
`com.google.mlkit.genai.common.GenAiException`	if initializing the inference engine fails.