About the Vercel AI SDK integration
telemetry.dev plugs into the Vercel AI SDK through OpenTelemetry. Enable telemetry on your calls, point the exporter at our endpoint, and every generateText, streamText, and tool call arrives as a span — complete with prompts, completions, token counts, and timing.
Key features
- Drop-in setup: Install
@telemetry-dev/ai-sdk, add your API key, and you're sending traces — no manual spans.
- Full request traces: Prompts, completions, tool calls, and streamed responses stitched into one trace.
- Tokens and cost: Cost is computed server-side from real token usage and current model pricing.
- Group by anything: Break latency and spend down by model, provider, environment, or user.
- Standard OTLP, no lock-in: Built on OpenTelemetry, so your data stays portable.
Get started
Install the package, set your TELEMETRY_DEV_API_KEY, and turn on experimental_telemetry in your AI SDK calls. Your first traces land in about five minutes.