telemetry.dev

About

An observability platform for AI and LLM applications that records every model call, tool step, and retrieval as a trace with tokens, cost, latency, and errors attached. It is OpenTelemetry-native, so traces arrive either through first-party TypeScript and Python SDKs (including a Vercel AI SDK integration) or any standard OTLP/HTTP exporter, and cost is computed per call from live pricing for 4,700+ models. Captured inputs and outputs are opt-in per environment with secret redaction at ingest, and traces can be searched and grouped by model, provider, environment, or user.