Function generateEmbeddings

apps/sim/lib/knowledge/embeddings.ts:353–394

(
  texts: string[],
  embeddingModel: string = DEFAULT_EMBEDDING_MODEL,
  workspaceId?: string | null
)

 * Generate embeddings for multiple texts with token-aware batching and parallel processing.
 */
export async function generateEmbeddings(
  texts: string[],
  embeddingModel: string = DEFAULT_EMBEDDING_MODEL,
  workspaceId?: string | null
): Promise<GenerateEmbeddingsResult> {
  const provider = await resolveProvider(embeddingModel, workspaceId)

  const tokenBatches = batchByTokenLimit(texts, MAX_TOKENS_PER_REQUEST, embeddingModel)
  const batches = provider.maxItemsPerRequest
    ? tokenBatches.flatMap((batch) => splitByItemLimit(batch, provider.maxItemsPerRequest!))
    : tokenBatches

  const batchResults = await processWithConcurrency(
    batches,
    MAX_CONCURRENT_BATCHES,
    async (batch, i) => {
      try {
        return await callEmbeddingAPI(batch, provider, 'document')
      } catch (error) {
        logger.error(`Failed to generate embeddings for batch ${i + 1}/${batches.length}:`, error)
        throw error
      }
    }
  )

  const allEmbeddings: number[][] = []
  let totalTokens = 0
  for (const batch of batchResults) {
    for (const emb of batch.embeddings) {
      allEmbeddings.push(emb)
    }
    totalTokens += batch.totalTokens
  }

  return {
    embeddings: allEmbeddings,
    totalTokens,
    isBYOK: provider.isBYOK,
    modelName: provider.modelName,
    pricingId: provider.pricingId,
  }
}

/**
 * Generate embedding for a single search query.