feat(embedding): surface token usage from AI SDK

Wire up usage.tokens from embed/embedMany return values so callers
can track actual token consumption for billing and observability.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jgpruitt

packages/embedding/generate.test.ts

@@ -22,11 +22,12 @@ const ollamaConfig: EmbeddingConfig = {
 
 describe.skipIf(!RUN_INTEGRATION)("embedding integration (ollama)", () => {
   test("generateEmbedding returns correct dimensions", async () => {
-    const embedding = await generateEmbedding("test text", ollamaConfig);
+    const result = await generateEmbedding("test text", ollamaConfig);
 
-    expect(embedding).toBeInstanceOf(Array);
-    expect(embedding.length).toBe(768);
-    expect(typeof embedding[0]).toBe("number");
+    expect(result.embedding).toBeInstanceOf(Array);
+    expect(result.embedding.length).toBe(768);
+    expect(typeof result.embedding[0]).toBe("number");
+    expect(result.tokens).toBeGreaterThan(0);
   });
 
   test("generateEmbedding handles long text with truncation", async () => {
@@ -36,8 +37,9 @@ describe.skipIf(!RUN_INTEGRATION)("embedding integration (ollama)", () => {
       options: { maxTokens: 8000 },
     };
 
-    const embedding = await generateEmbedding(longText, configWithTruncation);
-    expect(embedding.length).toBe(768);
+    const result = await generateEmbedding(longText, configWithTruncation);
+    expect(result.embedding.length).toBe(768);
+    expect(result.tokens).toBeGreaterThan(0);
   });
 
   test("generateEmbeddings returns results for batch", async () => {

packages/embedding/generate.ts

@@ -48,6 +48,12 @@ function prepareText(text: string, config: EmbeddingConfig): string {
 // Single Embedding
 // =============================================================================
 
+export interface SingleEmbedResult {
+  embedding: number[];
+  /** Tokens consumed by the embedding API call */
+  tokens: number;
+}
+
 /**
  * Generate a single embedding for text.
  *
@@ -58,7 +64,7 @@ function prepareText(text: string, config: EmbeddingConfig): string {
 export async function generateEmbedding(
   text: string,
   config: EmbeddingConfig,
-): Promise<number[]> {
+): Promise<SingleEmbedResult> {
   const model = getEmbeddingModel(config);
   const preparedText = prepareText(text, config);
 
@@ -74,7 +80,10 @@ export async function generateEmbedding(
     );
   }
 
-  return result.embedding;
+  return {
+    embedding: result.embedding,
+    tokens: result.usage.tokens,
+  };
 }
 
 // =============================================================================
@@ -107,12 +116,15 @@ export async function generateEmbeddings(
 
   try {
     // Try batch API first
-    const { embeddings } = await embedMany({
+    const { embeddings, usage } = await embedMany({
       model,
       values: texts,
       ...getEmbedOptions(config),
     });
 
+    // embedMany returns aggregate token count — distribute evenly
+    const tokensPerRow = Math.floor(usage.tokens / rows.length);
+
     for (let i = 0; i < rows.length; i++) {
       const embedding = embeddings[i];
       const row = rows[i];
@@ -128,6 +140,7 @@ export async function generateEmbeddings(
         results.push({
           id: row.id,
           embedding,
+          tokens: tokensPerRow,
         });
       }
     }
@@ -155,6 +168,7 @@ export async function generateEmbeddings(
           results.push({
             id: row.id,
             embedding: result.embedding,
+            tokens: result.usage.tokens,
           });
         }
       } catch (err) {

packages/embedding/index.ts

@@ -20,6 +20,7 @@ export {
   generateEmbedding,
   generateEmbeddings,
   validateConfig,
+  type SingleEmbedResult,
 } from "./generate";
 
 // =============================================================================

packages/embedding/types.ts

@@ -51,6 +51,8 @@ export interface MemoryRow {
 export interface EmbedResult {
   id: string;
   embedding: number[];
+  /** Tokens consumed by the embedding API call */
+  tokens?: number;
   error?: string;
 }