Installation

The AI SDK integration traces generateText, streamText, and other Vercel AI SDK calls automatically.

Install

npm install @breadcrumb-sdk/core @breadcrumb-sdk/ai-sdk

Setup

import { init } from "@breadcrumb-sdk/core";
import { initAiSdk } from "@breadcrumb-sdk/ai-sdk";

const bc = init({
  apiKey: process.env.BREADCRUMB_API_KEY!,
  baseUrl: process.env.BREADCRUMB_BASE_URL!,
});

const { telemetry } = initAiSdk(bc);

Usage

Pass the telemetry helper to any AI SDK call:

const { text } = await generateText({
  model: anthropic("claude-sonnet-4-5"),
  prompt: "Summarize this document.",
  experimental_telemetry: telemetry("summarize"),
});

The string you pass to telemetry() becomes the trace name in the dashboard.

Grouping calls

Wrap multiple AI SDK calls in a single trace using bc.trace():

await bc.trace("multi-step", async (root) => {

  const { text: plan } = await generateText({
    model: anthropic("claude-sonnet-4-5"),
    prompt: "Plan the steps to answer this question.",
    experimental_telemetry: telemetry("plan"),
  });

  const { text: answer } = await generateText({
    model: anthropic("claude-sonnet-4-5"),
    prompt: `Execute this plan: ${plan}`,
    experimental_telemetry: telemetry("execute"),
  });

});

Each generateText call appears as a nested span inside the parent trace.

Tools and agents

Tool calls and multi-step agent loops are traced automatically. Each tool invocation appears as a nested span inside the trace.

Using with Sentry, Langfuse, or other OTel tools

Breadcrumb works alongside other OpenTelemetry tools without configuration. See the compatibility guide for details on shared providers and custom tracers.