Missing input/output on trace when using openai js integration

[dimitrydesilias]

Hi Support,
I have a problem with my tracing details
it returns null data ,
I am working on a next js project with vercel ai sdk I have changed the export const openAiClient = observeOpenAI(new OpenAI({ apiKey: env.OPENAI_API_KEY, })); I received null for the input to output do you have any idea where I missed something

[dosubot[bot]]

Hey @dimitrydesilias! I'm here to help you with your tracing details. I can assist with bug fixes, answer questions, and guide you as a contributor. Let's tackle this together!

There are a few common issues and solutions when integrating Langfuse with the Vercel AI SDK that might help resolve the problem of null data in your tracing details:

1. Enable Telemetry: Ensure telemetry is enabled on each request you want to trace.

   const result = await generateText({
     model: openai("gpt-4-turbo"),
     prompt: "Write a short story about a cat.",
     experimental_telemetry: { isEnabled: true },
   });

2. Correct SDK Version: Make sure you are using "ai": "^3.3.0" or later.

   "dependencies": {
     "ai": "^3.3.0"
   }

3. Set Up LangfuseExporter: Ensure you have added the LangfuseExporter to your application and set the necessary environment variables.

   LANGFUSE_SECRET_KEY="sk-lf-..."
   LANGFUSE_PUBLIC_KEY="pk-lf-..."
   LANGFUSE_BASEURL="https://cloud.langfuse.com"

4. Register the Exporter: Register the LangfuseExporter via the OpenTelemetry SDK.

   import { registerOTel } from "@vercel/otel";
   import { LangfuseExporter } from "langfuse-vercel";
   
   export function register() {
     registerOTel({
       serviceName: "langfuse-vercel-ai-nextjs-example",
       traceExporter: new LangfuseExporter(),
     });
   }

5. Enable Instrumentation Hook: Ensure the instrumentationHook is enabled in your next.config.js.

   /** @type {import('next').NextConfig} */
   const nextConfig = {
     experimental: {
       instrumentationHook: true,
     },
   };
   
   module.exports = nextConfig;

Additionally, enabling the debug option can help you see the logs of the exporter for further troubleshooting:

new LangfuseExporter({ debug: true });

For more detailed steps and examples, you can refer to the Langfuse documentation on integrating with the Vercel AI SDK ^[1][2].

To continue talking to Dosu, mention @dosu.

[marcklingen]

Can you check whether you see any input/output in the generation (select within the tree view on the right)

[hassiebp]

@dimitrydesilias If you are using the Langfuse Vercel AI SDK integration and rely on the AI SDK functions (streamText, generateText etc) you won't need to additionally wrap the OpenAI client with the Langfuse observeOpenAI wrapper. Is there a specific reason you chose to do so?

[dimitrydesilias]

Hello @hassiebp, When I tried to enable telemetry I get this error.
version ai 3.4.7

[hassiebp]

@dimitrydesilias this is odd. Could you please try restarting your TS server? If you have followed the AI SDK docs here, then please open an issue in their repo with regards to this type error 🙏🏾

[dimitrydesilias]

Hi @hassiebp,

Does Langfuse work with the OpenAI Assistant? On our side, we don't use streamText and generateText, but rather the OpenAI helper. This is also one of the reasons why I'm trying to wrap the OpenAI instance, which returns null data.

[hassiebp]

@dimitrydesilias Apologies, but I'm not following - what do you mean specifically with OpenAI Assistant? The screenshot you shared above is using streamText though from the Vercel AI SDK? Could you please share a full repro example of your issue?