feat(node): Add OpenAI integration #17022

RulaKhaled · 2025-07-15T15:02:25Z

This PR adds official support for instrumenting OpenAI SDK calls in Node with Sentry tracing, following OpenTelemetry semantic conventions for Generative AI.

We currently instrument the following OpenAI SDK methods:

client.chat.completions.create() - For chat-based completions
client.responses.create() - For the responses API

Currently supported:

Node.js - Mentioned methods are supported in this PR
ESM and CJS - Both module systems are supported

The openAIIntegration() accepts the following options:

// The integration respects your sendDefaultPii client option
interface OpenAiOptions {
  recordInputs?: boolean;   // Whether to record prompt messages
  recordOutputs?: boolean;  // Whether to record response text
}

Example:

Sentry.init({
  dsn: '__DSN__',
  sendDefaultPii: false, // Even with PII disabled globally
  integrations: [
    Sentry.openAIIntegration({
      recordInputs: true,    // Force recording prompts
      recordOutputs: true,   // Force recording responses
    }),
  ],
});

github-actions · 2025-07-15T15:10:42Z

size-limit report 📦

Path	Size	% Change	Change
@sentry/browser	23.78 kB	-	-
@sentry/browser - with treeshaking flags	22.34 kB	-	-
@sentry/browser (incl. Tracing)	39.66 kB	-	-
@sentry/browser (incl. Tracing, Replay)	77.79 kB	-	-
@sentry/browser (incl. Tracing, Replay) - with treeshaking flags	67.59 kB	-	-
@sentry/browser (incl. Tracing, Replay with Canvas)	82.49 kB	-	-
@sentry/browser (incl. Tracing, Replay, Feedback)	94.59 kB	-	-
@sentry/browser (incl. Feedback)	40.48 kB	-	-
@sentry/browser (incl. sendFeedback)	28.46 kB	-	-
@sentry/browser (incl. FeedbackAsync)	33.36 kB	-	-
@sentry/react	25.54 kB	-	-
@sentry/react (incl. Tracing)	41.62 kB	-	-
@sentry/vue	28.23 kB	-	-
@sentry/vue (incl. Tracing)	41.45 kB	-	-
@sentry/svelte	23.81 kB	-	-
CDN Bundle	25.18 kB	-	-
CDN Bundle (incl. Tracing)	39.42 kB	-	-
CDN Bundle (incl. Tracing, Replay)	75.42 kB	-	-
CDN Bundle (incl. Tracing, Replay, Feedback)	80.89 kB	-	-
CDN Bundle - uncompressed	73.44 kB	-	-
CDN Bundle (incl. Tracing) - uncompressed	116.86 kB	-	-
CDN Bundle (incl. Tracing, Replay) - uncompressed	231 kB	-	-
CDN Bundle (incl. Tracing, Replay, Feedback) - uncompressed	243.81 kB	-	-
@sentry/nextjs (client)	43.66 kB	-	-
@sentry/sveltekit (client)	40.08 kB	-	-
@sentry/node	168.98 kB	+0.7%	+1.17 kB 🔺
@sentry/node - without tracing	100.2 kB	-	-
@sentry/aws-serverless	128.36 kB	-	-

View base workflow run

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

AbhiPrasad

Do the types and utils for openai have to live in @sentry/core? It's node only, I'd rather that they are in the node package.

The only thing that should be in core are the openai-attributes. I would prefer if we renamed that file to gen-ai-attributes.ts and made it more generic (the vercel ai attributes file can just re-export the attributes from gen-ai-attributes.ts).

We put the vercelai stuff in @sentry/core because it can also be used by the cloudflare and vercel-edge sdks.

RulaKhaled · 2025-07-16T13:31:17Z

cloudflare and vercel-edge sdks

Do the types and utils for openai have to live in @sentry/core? It's node only, I'd rather that they are in the node package.

The only thing that should be in core are the openai-attributes. I would prefer if we renamed that file to gen-ai-attributes.ts and made it more generic (the vercel ai attributes file can just re-export the attributes from gen-ai-attributes.ts).

We put the vercelai stuff in @sentry/core because it can also be used by the cloudflare and vercel-edge sdks.

The types and utils are primarily used in openai.ts which is the main file we’ll use to instrument openai in both Vercel Edge and Cloudflare. We’re reusing the types defined in core within the Node integration, which is better than redefining them in Node (something i noticed we do for vercel ai), and potentially again in other places where we want to support openai.

Sure thing, I will rename it to gen-ai-attributes.ts so we can make use of it later

AbhiPrasad · 2025-07-16T13:40:24Z

We’re reusing the types defined in core within the Node integration, which is better than redefining them in Node (something i noticed we do for vercel ai), and potentially again in other places where we want to support openai.

we're using import-in-the-middle to instrument this package, and openai sdk itself only works on the node runtime, so I think having it in core is optimizing too early. It'll take us some time to figure out non-node runtimes, once we do we can easily move stuff from node -> core. For now I'd like us to have the logic in the node sdk.

RulaKhaled · 2025-07-16T13:50:51Z

We’re reusing the types defined in core within the Node integration, which is better than redefining them in Node (something i noticed we do for vercel ai), and potentially again in other places where we want to support openai.

we're using import-in-the-middle to instrument this package, and openai sdk itself only works on the node runtime, so I think having it in core is optimizing too early. It'll take us some time to figure out non-node runtimes, once we do we can easily move stuff from node -> core. For now I'd like us to have the logic in the node sdk.

Oh great call, didn't really know openai sdk itself only works on the node runtime, will move!

andreiborza · 2025-07-16T13:55:11Z

@AbhiPrasad are you sure about the library not running on non-node runtimes? According to their README vercel-edge and cloudflare is supported.

We discussed on Monday in person that we'll start with this being a node instrumentation but that we want to have it available on vercel-edge and cloudflare eventually. We can't use import-in-the-middle, but we discussed that users might have to instrument their open ai client manually by calling a function we export in those runtimes.

AbhiPrasad · 2025-07-16T14:05:40Z

@AbhiPrasad are you sure about the library not running on non-node runtimes? According to their README vercel-edge and cloudflare is supported.

This is my mistake! looks like with the latest major version they don't rely on node types for anything anymore (https://github.com/openai/openai-node/blob/2d0c448912f0adc1caa5052937971bc7bd6daaee/MIGRATION.md?plain=1#L23-L26). Sorry for the confusion.

I still think this should live in @sentry/node until we've figured out the alternative API to using import-in-the-middle. I would prefer to constrain the surface area.

RulaKhaled · 2025-07-17T07:59:32Z

@AbhiPrasad are you sure about the library not running on non-node runtimes? According to their README vercel-edge and cloudflare is supported.

This is my mistake! looks like with the latest major version they don't rely on node types for anything anymore (https://github.com/openai/openai-node/blob/2d0c448912f0adc1caa5052937971bc7bd6daaee/MIGRATION.md?plain=1#L23-L26). Sorry for the confusion.

I still think this should live in @sentry/node until we've figured out the alternative API to using import-in-the-middle. I would prefer to constrain the surface area.

Just a follow up on this, after syncing with Abhi, we have decided to keep things in core for now as we are aiming to tackle non-node environment very soon

packages/core/src/utils/openai-constants.ts

packages/node/src/integrations/tracing/openai/instrumentation.ts

mydea · 2025-07-17T08:23:14Z

packages/node/src/integrations/tracing/openai/instrumentation.ts

+      });
+    } as unknown as abstract new (...args: unknown[]) => OpenAiClient;
+
+    // Preserve static and prototype chains


l: I wonder if we can extract this into a util and re-use this, as I guess this is more boilerplate at this point. Nothing to do in this PR, this is fine, just maybe a future iteration improvement.

I agree. I think I'll add that to the refactor we're planning for shared utils and attributes for vercel AI and openai

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

packages/core/src/index.ts

packages/core/src/utils/openai-utils.ts

packages/node/src/integrations/tracing/openai/instrumentation.ts

cursor

Bug: OpenAI Integration Logic and Data Handling Errors

The OpenAI instrumentation has three issues:

The logic for shouldRecordInputsAndOutputs incorrectly defaults to false if the OpenAI integration is not found, ignoring the client's sendDefaultPii setting.
The optional chaining for client?.getOptions().sendDefaultPii is incomplete, potentially causing a TypeError if getOptions() returns undefined.
In addRequestAttributes, both params.messages and params.input set the same GEN_AI_REQUEST_MESSAGES_ATTRIBUTE, leading to data loss if both are present in the request.

packages/core/src/utils/openai/index.ts#L191-L204

sentry-javascript/packages/core/src/utils/openai/index.ts

Lines 191 to 204 in 86927ff

    
           function addRequestAttributes(span: Span, params: Record<string, unknown>): void { 
        
             if ('messages' in params) { 
        
               span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.messages) }); 
        
             } 
        
             if ('input' in params) { 
        
               span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.input) }); 
        
             } 
        
           } 
        
           function getOptionsFromIntegration(): OpenAiOptions { 
        
             const scope = getCurrentScope(); 
        
             const client = scope.getClient(); 
        
             const integration = client?.getIntegrationByName(OPENAI_INTEGRATION_NAME) as OpenAiIntegration | undefined; 
        
             const shouldRecordInputsAndOutputs = integration ? Boolean(client?.getOptions().sendDefaultPii) : false;

Fix in Cursor • Fix in Web

Was this report helpful? Give feedback by reacting with 👍 or 👎

andreiborza

🚀 nice work!

feat: Instrument openai for node

c6e8561

RulaKhaled added 2 commits July 16, 2025 11:22

add tests, refactor some types

6966214

fix responses test

247bf9f

RulaKhaled changed the title ~~[WIP] feat(node): Instrument openai chat and responses api~~ feat(node): Add OpenAI integration Jul 16, 2025

quick refactor to record inputs option

8535978

RulaKhaled marked this pull request as ready for review July 16, 2025 10:47

This comment was marked as outdated.

Sign in to view

update name to reference integration name constant

ca2d43e

mydea reviewed Jul 16, 2025

View reviewed changes

dev-packages/node-integration-tests/suites/tracing/openai/scenario.mjs Outdated Show resolved Hide resolved

RulaKhaled requested review from AbhiPrasad and andreiborza July 16, 2025 13:10

AbhiPrasad reviewed Jul 16, 2025

View reviewed changes

rename to gen-ai-attributes, and fix test

db18c0a

mydea reviewed Jul 17, 2025

View reviewed changes

packages/core/src/utils/openai-constants.ts Outdated Show resolved Hide resolved

mydea reviewed Jul 17, 2025

View reviewed changes

packages/node/src/integrations/tracing/openai/instrumentation.ts Outdated Show resolved Hide resolved

mydea reviewed Jul 17, 2025

View reviewed changes

packages/node/src/integrations/tracing/openai/instrumentation.ts Outdated Show resolved Hide resolved

mydea reviewed Jul 17, 2025

View reviewed changes

Move stuff to sub folder, update with small refactor

e9d9f51

andreiborza reviewed Jul 17, 2025

View reviewed changes

RulaKhaled added 2 commits July 17, 2025 12:03

Remove vercel code leftovers from openai's

04a2dde

fix tests

86927ff

cursor bot reviewed Jul 17, 2025

View reviewed changes

andreiborza approved these changes Jul 17, 2025

View reviewed changes

andreiborza merged commit f538ef0 into develop Jul 17, 2025
650 of 659 checks passed

andreiborza deleted the openai-node branch July 17, 2025 11:34

This was referenced Jul 24, 2025

OpenAI integration is breaking openai calls, even when the integration isn't enabled #17159

Closed

chore(deps): downgrade sentry/node owid/owid-grapher#5184

Merged

	function addRequestAttributes(span: Span, params: Record<string, unknown>): void {
	if ('messages' in params) {
	span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.messages) });
	}
	if ('input' in params) {
	span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.input) });
	}
	}

	function getOptionsFromIntegration(): OpenAiOptions {
	const scope = getCurrentScope();
	const client = scope.getClient();
	const integration = client?.getIntegrationByName(OPENAI_INTEGRATION_NAME) as OpenAiIntegration \| undefined;
	const shouldRecordInputsAndOutputs = integration ? Boolean(client?.getOptions().sendDefaultPii) : false;

Uh oh!

feat(node): Add OpenAI integration #17022

feat(node): Add OpenAI integration #17022

Uh oh!

Conversation

RulaKhaled commented Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

size-limit report 📦

Uh oh!

This comment was marked as outdated.

Uh oh!

Uh oh!

AbhiPrasad left a comment

Choose a reason for hiding this comment

Uh oh!

RulaKhaled commented Jul 16, 2025

Uh oh!

AbhiPrasad commented Jul 16, 2025

Uh oh!

RulaKhaled commented Jul 16, 2025

Uh oh!

andreiborza commented Jul 16, 2025

Uh oh!

AbhiPrasad commented Jul 16, 2025

Uh oh!

RulaKhaled commented Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mydea Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

RulaKhaled Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Bug: OpenAI Integration Logic and Data Handling Errors

Uh oh!

andreiborza left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

RulaKhaled commented Jul 15, 2025 •

edited

Loading

github-actions bot commented Jul 15, 2025 •

edited

Loading

RulaKhaled commented Jul 17, 2025 •

edited

Loading