Observability for Vercel AI SDK with Opik

Setup

The AI SDK supports tracing via OpenTelemetry. With the OpikExporter you can collect these traces in Opik. While telemetry is experimental (docs), you can enable it by setting experimental_telemetry on each request that you want to trace.

const result = await generateText({
  model: openai("gpt-4o"),
  prompt: "Tell a joke",
  experimental_telemetry: { isEnabled: true },
});

To collect the traces in Opik, you need to add the OpikExporter to your application, first you have to set your environment variables

Intent: Use OpikExporter as the OTEL trace exporter for AI SDK telemetry.

Applies when: You are using generateText/streamText with experimental_telemetry.

Required fields:

• OPIK_API_KEY

Optional fields:

• OPIK_WORKSPACE (defaults to default for Cloud/Enterprise; set to override)
• OPIK_PROJECT_NAME (defaults to "Default Project"; set to override)
• OPIK_URL_OVERRIDE (required for non-default deployment URLs)
• OPIK_LOG_LEVEL

Deployment endpoint examples:

• Opik Cloud: OPIK_URL_OVERRIDE=https://www.comet.com/opik/api
• Enterprise: OPIK_URL_OVERRIDE=https://<comet-deployment-url>/opik/api
• Self-hosted: OPIK_URL_OVERRIDE=http://localhost:5173/api

OPIK_API_KEY="<opik-api-key>"
OPIK_URL_OVERRIDE=https://www.comet.com/opik/api # in case you are using the Cloud version
OPIK_PROJECT_NAME="<custom-project-name>"
OPIK_WORKSPACE="<your-workspace>"
OPENAI_API_KEY="<openai-api-key>" # in case you are using an OpenAI model

import { OpikExporter } from "opik-vercel";

new OpikExporter();

Now you need to register this exporter via the OpenTelemetry SDK.

Next.js

Next.js has support for OpenTelemetry instrumentation on the framework level. Learn more about it in the Next.js OpenTelemetry guide.

Install dependencies:

npm install opik-vercel @vercel/otel @opentelemetry/api-logs @opentelemetry/instrumentation @opentelemetry/sdk-logs

Add OpikExporter to your instrumentation file:

import { registerOTel } from "@vercel/otel";
import { OpikExporter } from "opik-vercel";

export function register() {
  registerOTel({
    serviceName: "opik-vercel-ai-nextjs-example",
    traceExporter: new OpikExporter(),
  });
}

Node.js

Install dependencies:

npm install opik-vercel ai @ai-sdk/openai @opentelemetry/sdk-node @opentelemetry/auto-instrumentations-node

import { openai } from "@ai-sdk/openai";
import { generateText } from "ai";
import { NodeSDK } from "@opentelemetry/sdk-node";
import { getNodeAutoInstrumentations } from "@opentelemetry/auto-instrumentations-node";
import { OpikExporter } from "opik-vercel";

const sdk = new NodeSDK({
  traceExporter: new OpikExporter(),
  instrumentations: [getNodeAutoInstrumentations()],
});

sdk.start();

async function main() {
  const result = await generateText({
    model: openai("gpt-4o"),
    maxTokens: 50,
    prompt: "What is love?",
    experimental_telemetry: OpikExporter.getSettings({
      name: "opik-nodejs-example",
    }),
  });

  console.log(result.text);

  await sdk.shutdown(); // Flushes the trace to Opik
}

main().catch(console.error);

Done! All traces that contain AI SDK spans are automatically captured in Opik.

AI SDK versions and Vercel eve

OpikExporter works with all current AI SDK versions — v4, v5, v6, and v7 — with no extra setup. It also supports Vercel eve, the agent framework built on AI SDK v7, capturing each run as nested LLM and tool spans. The setup shown above is all you need, and your code keeps working as you upgrade.

Multi-turn eve conversations

When you run Vercel eve, a multi-turn conversation is grouped into a single Opik thread automatically, so every turn shows up together — you don't need to set threadId yourself.

Cached token usage

If your model reports prompt caching, cached tokens are tracked alongside the usual input and output tokens, so your token and cost metrics stay accurate.

Configuration

Custom Tags and Metadata

You can add custom tags and metadata to all traces generated by the OpikExporter:

const exporter = new OpikExporter({
  // Optional: add custom tags to all traces
  tags: ["production", "gpt-4o"],
  // Optional: add custom metadata to all traces
  metadata: {
    environment: "production",
    version: "1.0.0",
    team: "ai-team",
  },
  // Optional: associate traces with a conversation thread
  threadId: "conversation-123",
});

Tags are useful for filtering and grouping traces, while metadata adds additional context that can be valuable for debugging and analysis. The threadId parameter is useful for tracking multi-turn conversations or grouping related AI interactions.

Pass Custom Trace name

const result = await generateText({
  model: openai("gpt-4o"),
  prompt: "Tell a joke",
  experimental_telemetry: OpikExporter.getSettings({
    name: "custom-trace-name",
  }),
});

Thread ID Support

You can associate traces with conversation threads by setting the threadId parameter. This is useful for tracking multi-turn conversations or grouping related AI interactions.

Set threadId per request via telemetry metadata (this overrides any exporter-level threadId):

const result = await generateText({
  model: openai("gpt-4o"),
  prompt: "Continue the conversation",
  experimental_telemetry: OpikExporter.getSettings({
    name: "chat-message",
    metadata: {
      threadId: "conversation-456",
    },
  }),
});

Debugging

Use the logger level to see the more verbose logs of the exporter.

OPIK_LOG_LEVEL=DEBUG

Validation

1. Run one AI SDK request with experimental_telemetry enabled.
2. Confirm OpikExporter initializes without auth errors.
3. Verify traces in the target Opik workspace/project.

Source references

• Vercel AI SDK telemetry
• Next.js OpenTelemetry guide
• Opik Vercel exporter package