feat(core): Support embeddings in langchain (#20017)

Adds embeddings instrumentation for LangChain (`embedQuery`,
`embedDocuments`). LangChain's `Embeddings` base class doesn't use the
callback system that we rely on for chat models, so this uses direct
prototype patching instead.

In Node.js, embedding classes from provider packages are
auto-instrumented. For other runtimes:
   
```js
  const embeddings = Sentry.instrumentLangChainEmbeddings(
    new OpenAIEmbeddings({ model: 'text-embedding-3-small' })
);
  await embeddings.embedQuery('Hello world');
```
                                                                                                                              
Creates `gen_ai.embeddings` spans. Token usage is not available since LangChain's embedding methods only return raw vectors.

Closes #19774

Author: nicohrubec

CHANGELOG.md

@@ -23,6 +23,19 @@
 
   If you reference these attributes in hooks (e.g. `beforeSendTransaction`), update them to the `gen_ai.*` equivalents.
 
+- feat(langchain): Support embeddings APIs in LangChain ([#20017](https://github.com/getsentry/sentry-javascript/pull/20017))
+
+  Adds instrumentation for LangChain embeddings (`embedQuery`, `embedDocuments`), creating `gen_ai.embeddings` spans. In Node.js, embedding classes from `@langchain/openai`, `@langchain/google-genai`, `@langchain/mistralai`, and `@langchain/google-vertexai` are auto-instrumented. For other runtimes, use the new `instrumentLangChainEmbeddings` API:
+
+  ```javascript
+  import * as Sentry from '@sentry/cloudflare';
+  import { OpenAIEmbeddings } from '@langchain/openai';
+
+  const embeddings = Sentry.instrumentLangChainEmbeddings(new OpenAIEmbeddings({ model: 'text-embedding-3-small' }));
+
+  await embeddings.embedQuery('Hello world');
+  ```
+
 ## 10.47.0
 
 ### Important Changes

dev-packages/browser-integration-tests/suites/tracing/ai-providers/langchain/mocks.js

@@ -1,3 +1,21 @@
+// Mock LangChain Embeddings for browser testing
+export class MockOpenAIEmbeddings {
+  constructor(params) {
+    this.model = params.model;
+    this.dimensions = params.dimensions;
+  }
+
+  async embedQuery(_text) {
+    await new Promise(resolve => setTimeout(resolve, 10));
+    return [0.1, 0.2, 0.3];
+  }
+
+  async embedDocuments(documents) {
+    await new Promise(resolve => setTimeout(resolve, 10));
+    return documents.map(() => [0.1, 0.2, 0.3]);
+  }
+}
+
 // Mock LangChain Chat Model for browser testing
 export class MockChatAnthropic {
   constructor(params) {

dev-packages/browser-integration-tests/suites/tracing/ai-providers/langchain/subject.js

@@ -1,5 +1,6 @@
 import { createLangChainCallbackHandler } from '@sentry/browser';
-import { MockChatAnthropic } from './mocks.js';
+import { instrumentLangChainEmbeddings } from '@sentry/browser';
+import { MockChatAnthropic, MockOpenAIEmbeddings } from './mocks.js';
 
 const callbackHandler = createLangChainCallbackHandler({
   recordInputs: false,
@@ -20,3 +21,11 @@ const response = await chatModel.invoke('What is the capital of France?', {
 });
 
 console.log('Received response', response);
+
+// Test embeddings instrumentation
+const embeddings = instrumentLangChainEmbeddings(
+  new MockOpenAIEmbeddings({ model: 'text-embedding-3-small', dimensions: 1536 }),
+);
+
+const embedding = await embeddings.embedQuery('Hello world');
+console.log('Received embedding', embedding);

dev-packages/browser-integration-tests/suites/tracing/ai-providers/langchain/test.ts

@@ -35,3 +35,29 @@ sentryTest('manual LangChain instrumentation sends gen_ai transactions', async (
     'gen_ai.usage.total_tokens': 25,
   });
 });
+
+sentryTest(
+  'manual LangChain embeddings instrumentation sends gen_ai transactions',
+  async ({ getLocalTestUrl, page }) => {
+    const transactionPromise = waitForTransactionRequest(page, event => {
+      return !!event.transaction?.includes('text-embedding-3-small');
+    });
+
+    const url = await getLocalTestUrl({ testDir: __dirname });
+    await page.goto(url);
+
+    const req = await transactionPromise;
+
+    const eventData = envelopeRequestParser(req);
+
+    expect(eventData.transaction).toBe('embeddings text-embedding-3-small');
+    expect(eventData.contexts?.trace?.op).toBe('gen_ai.embeddings');
+    expect(eventData.contexts?.trace?.origin).toBe('auto.ai.langchain');
+    expect(eventData.contexts?.trace?.data).toMatchObject({
+      'gen_ai.operation.name': 'embeddings',
+      'gen_ai.system': 'openai',
+      'gen_ai.request.model': 'text-embedding-3-small',
+      'gen_ai.request.dimensions': 1536,
+    });
+  },
+);

dev-packages/browser-integration-tests/utils/generatePlugin.ts

@@ -42,6 +42,7 @@ const IMPORTED_INTEGRATION_CDN_BUNDLE_PATHS: Record<string, string> = {
   instrumentGoogleGenAIClient: 'instrumentgooglegenaiclient',
   instrumentLangGraph: 'instrumentlanggraph',
   createLangChainCallbackHandler: 'createlangchaincallbackhandler',
+  instrumentLangChainEmbeddings: 'instrumentlangchainembeddings',
   // technically, this is not an integration, but let's add it anyway for simplicity
   makeMultiplexedTransport: 'multiplexedtransport',
 };

dev-packages/node-integration-tests/package.json

@@ -32,6 +32,7 @@
     "@hono/node-server": "^1.19.10",
     "@langchain/anthropic": "^0.3.10",
     "@langchain/core": "^0.3.80",
+    "@langchain/openai": "^0.5.0",
     "@langchain/langgraph": "^0.2.32",
     "@nestjs/common": "^11",
     "@nestjs/core": "^11",

dev-packages/node-integration-tests/suites/tracing/langchain/instrument-with-pii.mjs

@@ -9,7 +9,7 @@ Sentry.init({
   transport: loggingTransport,
   beforeSendTransaction: event => {
     // Filter out mock express server transactions
-    if (event.transaction.includes('/v1/messages')) {
+    if (event.transaction.includes('/v1/messages') || event.transaction.includes('/v1/embeddings')) {
       return null;
     }
     return event;

dev-packages/node-integration-tests/suites/tracing/langchain/instrument.mjs

@@ -9,7 +9,7 @@ Sentry.init({
   transport: loggingTransport,
   beforeSendTransaction: event => {
     // Filter out mock express server transactions
-    if (event.transaction.includes('/v1/messages')) {
+    if (event.transaction.includes('/v1/messages') || event.transaction.includes('/v1/embeddings')) {
       return null;
     }
     return event;

dev-packages/node-integration-tests/suites/tracing/langchain/scenario-embeddings.mjs

@@ -0,0 +1,82 @@
+import { OpenAIEmbeddings } from '@langchain/openai';
+import * as Sentry from '@sentry/node';
+import express from 'express';
+
+function startMockOpenAIServer() {
+  const app = express();
+  app.use(express.json());
+
+  app.post('/v1/embeddings', (req, res) => {
+    const { model, input } = req.body;
+
+    if (model === 'error-model') {
+      res.status(400).json({
+        error: {
+          message: 'Model not found',
+          type: 'invalid_request_error',
+        },
+      });
+      return;
+    }
+
+    const inputs = Array.isArray(input) ? input : [input];
+    res.json({
+      object: 'list',
+      data: inputs.map((_, i) => ({
+        object: 'embedding',
+        embedding: [0.1, 0.2, 0.3],
+        index: i,
+      })),
+      model: model,
+      usage: {
+        prompt_tokens: 10,
+        total_tokens: 10,
+      },
+    });
+  });
+
+  return new Promise(resolve => {
+    const server = app.listen(0, () => {
+      resolve(server);
+    });
+  });
+}
+
+async function run() {
+  const server = await startMockOpenAIServer();
+  const baseUrl = `http://localhost:${server.address().port}/v1`;
+
+  await Sentry.startSpan({ op: 'function', name: 'main' }, async () => {
+    // Test 1: embedQuery
+    const embeddings = new OpenAIEmbeddings({
+      model: 'text-embedding-3-small',
+      dimensions: 1536,
+      apiKey: 'mock-api-key',
+      configuration: { baseURL: baseUrl },
+    });
+
+    await embeddings.embedQuery('Hello world');
+
+    // Test 2: embedDocuments
+    await embeddings.embedDocuments(['First document', 'Second document']);
+
+    // Test 3: Error handling
+    const errorEmbeddings = new OpenAIEmbeddings({
+      model: 'error-model',
+      apiKey: 'mock-api-key',
+      configuration: { baseURL: baseUrl },
+    });
+
+    try {
+      await errorEmbeddings.embedQuery('This will fail');
+    } catch {
+      // Expected error
+    }
+  });
+
+  await Sentry.flush(2000);
+
+  server.close();
+}
+
+run();

dev-packages/node-integration-tests/suites/tracing/langchain/test.ts

@@ -1,9 +1,12 @@
 import { SEMANTIC_ATTRIBUTE_SENTRY_OP, SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN } from '@sentry/core';
 import { afterAll, describe, expect } from 'vitest';
 import {
+  GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+  GEN_AI_EMBEDDINGS_INPUT_ATTRIBUTE,
   GEN_AI_INPUT_MESSAGES_ATTRIBUTE,
   GEN_AI_INPUT_MESSAGES_ORIGINAL_LENGTH_ATTRIBUTE,
   GEN_AI_OPERATION_NAME_ATTRIBUTE,
+  GEN_AI_REQUEST_DIMENSIONS_ATTRIBUTE,
   GEN_AI_REQUEST_MAX_TOKENS_ATTRIBUTE,
   GEN_AI_REQUEST_MODEL_ATTRIBUTE,
   GEN_AI_REQUEST_TEMPERATURE_ATTRIBUTE,
@@ -430,4 +433,120 @@ describe('LangChain integration', () => {
         .completed();
     });
   });
+
+  // =========================================================================
+  // Embeddings tests
+  // =========================================================================
+
+  const EXPECTED_TRANSACTION_EMBEDDINGS = {
+    transaction: 'main',
+    spans: expect.arrayContaining([
+      // embedQuery span
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_OPERATION_NAME_ATTRIBUTE]: 'embeddings',
+          [SEMANTIC_ATTRIBUTE_SENTRY_OP]: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+          [SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN]: 'auto.ai.langchain',
+          [GEN_AI_SYSTEM_ATTRIBUTE]: 'openai',
+          [GEN_AI_REQUEST_MODEL_ATTRIBUTE]: 'text-embedding-3-small',
+          [GEN_AI_REQUEST_DIMENSIONS_ATTRIBUTE]: 1536,
+        }),
+        description: 'embeddings text-embedding-3-small',
+        op: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+        origin: 'auto.ai.langchain',
+        status: 'ok',
+      }),
+      // embedDocuments span
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_OPERATION_NAME_ATTRIBUTE]: 'embeddings',
+          [SEMANTIC_ATTRIBUTE_SENTRY_OP]: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+          [SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN]: 'auto.ai.langchain',
+          [GEN_AI_SYSTEM_ATTRIBUTE]: 'openai',
+          [GEN_AI_REQUEST_MODEL_ATTRIBUTE]: 'text-embedding-3-small',
+        }),
+        description: 'embeddings text-embedding-3-small',
+        op: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+        origin: 'auto.ai.langchain',
+        status: 'ok',
+      }),
+      // Error span
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_OPERATION_NAME_ATTRIBUTE]: 'embeddings',
+          [SEMANTIC_ATTRIBUTE_SENTRY_OP]: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+          [SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN]: 'auto.ai.langchain',
+          [GEN_AI_SYSTEM_ATTRIBUTE]: 'openai',
+          [GEN_AI_REQUEST_MODEL_ATTRIBUTE]: 'error-model',
+        }),
+        description: 'embeddings error-model',
+        op: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+        origin: 'auto.ai.langchain',
+        status: 'internal_error',
+      }),
+    ]),
+  };
+
+  const EXPECTED_TRANSACTION_EMBEDDINGS_PII = {
+    transaction: 'main',
+    spans: expect.arrayContaining([
+      // embedQuery span with input recorded
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_OPERATION_NAME_ATTRIBUTE]: 'embeddings',
+          [SEMANTIC_ATTRIBUTE_SENTRY_OP]: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+          [SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN]: 'auto.ai.langchain',
+          [GEN_AI_SYSTEM_ATTRIBUTE]: 'openai',
+          [GEN_AI_REQUEST_MODEL_ATTRIBUTE]: 'text-embedding-3-small',
+          [GEN_AI_EMBEDDINGS_INPUT_ATTRIBUTE]: 'Hello world',
+        }),
+        description: 'embeddings text-embedding-3-small',
+        op: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+        origin: 'auto.ai.langchain',
+        status: 'ok',
+      }),
+      // embedDocuments span with input recorded
+      expect.objectContaining({
+        data: expect.objectContaining({
+          [GEN_AI_OPERATION_NAME_ATTRIBUTE]: 'embeddings',
+          [GEN_AI_EMBEDDINGS_INPUT_ATTRIBUTE]: JSON.stringify(['First document', 'Second document']),
+        }),
+        description: 'embeddings text-embedding-3-small',
+        op: GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE,
+        origin: 'auto.ai.langchain',
+        status: 'ok',
+      }),
+    ]),
+  };
+
+  createEsmAndCjsTests(__dirname, 'scenario-embeddings.mjs', 'instrument.mjs', (createRunner, test) => {
+    test('creates embedding spans with sendDefaultPii: false', async () => {
+      await createRunner().ignore('event').expect({ transaction: EXPECTED_TRANSACTION_EMBEDDINGS }).start().completed();
+    });
+
+    test('does not create duplicate embedding spans from double module patching', async () => {
+      await createRunner()
+        .ignore('event')
+        .expect({
+          transaction: event => {
+            const spans = event.spans || [];
+            const embeddingSpans = spans.filter(span => span.op === GEN_AI_EMBEDDINGS_OPERATION_ATTRIBUTE);
+            // The scenario makes 3 embedding calls (2 successful + 1 error).
+            expect(embeddingSpans).toHaveLength(3);
+          },
+        })
+        .start()
+        .completed();
+    });
+  });
+
+  createEsmAndCjsTests(__dirname, 'scenario-embeddings.mjs', 'instrument-with-pii.mjs', (createRunner, test) => {
+    test('creates embedding spans with sendDefaultPii: true', async () => {
+      await createRunner()
+        .ignore('event')
+        .expect({ transaction: EXPECTED_TRANSACTION_EMBEDDINGS_PII })
+        .start()
+        .completed();
+    });
+  });
 });