Skip to main content
Carbon turns every LLM and tool call your application makes into a queryable ledger. Each call is recorded as a structured event, and the dashboard slices that ledger by agent, model, user, and thread: which agents spend the most, which models are slow, which users drive usage, where errors cluster — down to the exact call.

​
How it works

1

Instrument your app

Wrap your existing OpenAI, Anthropic, or Vercel AI SDK client with the Carbon SDK. Your code keeps working as before; Carbon captures each call as an event in the background.
2

Events flow to Carbon

The SDK batches events and sends them to the ingest API, where they are validated, enriched with provider and cost data, and stored durably. Delivery is idempotent, so retries never create duplicates.
3

Explore the ledger

The dashboard slices your events by agent, model, user, thread, and tool over any time range — cost, usage, latency, and errors, from the top-level summary to the individual call.

​
Get started

Quickstart

Create an API key, install the SDK, and see your first events in a few minutes.

Integrations

Drop-in wrappers for the OpenAI, Anthropic, and Vercel AI SDKs.

Coding apps

Connect Claude Code and Codex in one command with @carbon-js/connect — no SDK.

Event model

What an event is, what it contains, and how traces tie events together.

API reference

Send events directly to the ingest API from any language.