Introduction

Breadcrumb is LLM tracing you install as a library, not a service you send your traces to. It runs in your backend, stores spans in a database you already own, and renders its dashboard as a component inside your app.

Three reasons to pick it over a hosted platform:

• Your traces sit next to your data. They are rows in your database, so nothing leaves your infrastructure and your own queries can join them to your users, tenants, and orders.
• The dashboard is a starting point. It ships as a React component you mount at your own route, so you can add pages to it, and when it stops fitting the way your product works, the same data is there as React hooks and typed queries.
• Nothing is metered. No account, no ingest endpoint, no per-event bill, so you can log the whole payload and keep it as long as you want.

How it works

Create one instance, pointed at your database.

import { breadcrumb } from "@breadcrumb-sh/core";
import { postgres } from "@breadcrumb-sh/core/adapters";

export const bc = breadcrumb({
  database: postgres(process.env.DATABASE_URL!),
  basePath: "/api/breadcrumb",
});

Mount bc.handler at that path and your traces are queryable. Render <BreadcrumbDashboard> at a route of your own, behind your own auth, and you can look at them:

<BreadcrumbDashboard api="/api/breadcrumb" basePath="/admin/traces" />

Pass bc.telemetry() to the Vercel AI SDK and your model calls start recording themselves. Everything else is optional: bc.trace() for code the SDK does not cover, and bc.api for querying traces yourself.

Next steps

• Quickstart: install, mount, and capture a trace.
• Frameworks: mount the handler in Next.js, Hono, or Node.
• Dashboard: mount the UI and add your own pages.
• Instrumenting: the AI SDK, OpenTelemetry, or manual spans.
• Database: adapters, schema, and migrations.
• Production: auth, retention, and going live.