Skip to main content
bryel is OpenTelemetry-native. The core is an OTLP/HTTP exporter plus a batching span processor that ship spans to the bryel ingest. Framework adapters add framework-aware mapping on top, promoting spans into the OpenInference convention before export. Pick the path that matches how you call your models.

Vercel AI SDK

Map the AI SDK’s OpenTelemetry spans into OpenInference with one processor and a per-call flag — models, tokens, cost, and the full multi-step trajectory.

Any OpenTelemetry app

Already emitting OpenInference spans? Add bryelSpanProcessor and you’re done — or supply a SpanMapper for custom span shapes.

Provider SDKs

Trace OpenAI, Anthropic, or LangChain by shipping their OpenTelemetry GenAI spans to bryel as-is — no call-site changes.
Adapters ship in their own packages — @bryel/vercel for the AI SDK — and build on the framework-agnostic @bryel/sdk-core. bryel owns its conventions outright, so no third-party observability SDK lands in your dependency tree.