Open-source observability for AI applications and agents. Trace model calls, inspect tool use, measure cost and latency, and find errors without changing providers.
A single request can call a model, run several tools, retry, and call the model again. telemetry.dev joins that work into one trace, so you can see the full path from input to output.
- Trace complete agent runs. See model calls, tool executions, retries, and errors in order.
- Measure usage and cost. Compare token use, latency, time to first token, and estimated cost.
- Inspect safely. Control input and output capture, mask sensitive values, and set content limits.
- Use your current stack. Send OpenTelemetry data or install an integration for your provider, framework, or coding agent.
Create a project at telemetry.dev, copy an API key, and install the SDK.
npm install @telemetry-dev/sdkimport { init } from "@telemetry-dev/sdk";
const telemetry = init({
apiKey: process.env.TELEMETRY_DEV_API_KEY,
});The SDK sends traces through OTLP, the OpenTelemetry Protocol. You can also use the Python SDK or an OpenTelemetry exporter.
telemetry.dev has integrations for OpenAI, Anthropic, Google Gen AI, Amazon Bedrock, LiteLLM, the Vercel AI SDK, TanStack AI, Eve, opencode, Oh My Pi, and Pi.
This repository contains the documentation for telemetry.dev. The product source is available in the telemetry.dev repository.