Skip to main content
The Vercel AI SDK includes built-in telemetry support that works seamlessly with Lemma’s OpenTelemetry-based tracing. This guide shows you how to set up tracing for your AI SDK applications.

​
How It Works

Once the tracer provider is registered, the AI SDK automatically captures spans for all generateText and streamText calls with telemetry enabled. Lemma receives:
  • Model calls and responses
  • Token usage (prompt tokens, completion tokens)
  • Tool invocations
  • Streaming events
  • Timing and latency

​
Getting Started

​
Install Dependencies

npm install @uselemma/tracing

​
Set Up the Tracer Provider

Create an instrumentation.ts (or instrumentation.js) file in your project root. Next.js will automatically run this code when your application starts: 
// instrumentation.ts
export async function register() {
  if (process.env.NEXT_RUNTIME === 'nodejs') {
    const { registerOTel } = await import('@uselemma/tracing');
    registerOTel();
  }
}
Make sure to enable instrumentation in your next.config.js: 
// next.config.js
module.exports = {
  experimental: {
    instrumentationHook: true,
  },
};

​
Node.js (Non-Next.js)

Create a tracer setup file and import it at the top of your entry point: 
// tracer.ts
import { registerOTel } from '@uselemma/tracing';
registerOTel();
Then import it at the top of your application entry point: 
// index.ts or server.ts
import './tracer'; // Must be first!
// ... rest of your imports

​
Options

registerOTel reads from LEMMA_API_KEY and LEMMA_PROJECT_ID environment variables by default. You can also pass options explicitly: 
registerOTel({
  apiKey: 'sk_...',
  projectId: '...',
  baseUrl: 'https://api.uselemma.ai', // default
});
Set the LEMMA_API_KEY and LEMMA_PROJECT_ID environment variables in your application. You can find these in your Lemma project settings.

​
Enable Telemetry in AI SDK

To capture traces from the Vercel AI SDK, enable telemetry in your streamText or generateText calls: 
import { streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

const result = await streamText({
  model: anthropic('claude-sonnet-4'),
  messages: [
    { role: 'user', content: 'What is the capital of France?' }
  ],
  experimental_telemetry: {
    isEnabled: true, // required
  },
});

​
Examples

​
Non-Streaming

import { wrapAgent } from "@uselemma/tracing";
import { generateText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

export const callAgent = async (userMessage: string) => {
  const wrappedFn = wrapAgent(
    "my-agent",
    async ({ onComplete }, input) => {
      const result = await generateText({
        model: anthropic('claude-sonnet-4'),
        messages: [{ role: 'user', content: input.userMessage }],
        experimental_telemetry: {
          isEnabled: true,
        },
      });

      onComplete(result.text);
      return result.text;
    }
  );

  const { result, runId } = await wrappedFn({ userMessage });
  return { result, runId };
};

​
Streaming

For streaming responses, set autoEndRoot: true so the RunBatchSpanProcessor automatically ends the root span when all direct child spans have finished. Call onComplete to record the output once the stream finishes: 
import { wrapAgent } from "@uselemma/tracing";
import { streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

export const callAgent = async (userMessage: string) => {
  const wrappedFn = wrapAgent(
    "my-agent",
    async ({ onComplete, recordError }, input) => {
      return streamText({
        model: anthropic('claude-sonnet-4'),
        messages: [{ role: 'user', content: input.userMessage }],
        experimental_telemetry: {
          isEnabled: true,
        },
        onFinish: (result) => {
          onComplete(result);
        },
        onError: recordError,
      });
    },
    { autoEndRoot: true }
  );

  const { result, runId } = await wrappedFn({ userMessage });
  return { result, runId };
};

​
Next Steps