OpenTelemetry GenAI.
The OpenTelemetry GenAI Semantic Conventions are the open standard for instrumenting LLM calls and AI agent runs. Trefur reads them natively — no proprietary mapping, no vendor lock-in.
What OpenTelemetry GenAI covers.
The OpenTelemetry GenAI Semantic Conventions define a vendor-neutral span and attribute shape for generative AI workloads. LLM calls — the prompt, completion, model name, token counts, latency, finish reason. Embedding calls — the input dimensions, the model, the latency. Tool calls — the tool name, arguments, result, status. Agent runs — the root span and every child span underneath.
The conventions are an open standard maintained by the OpenTelemetry community alongside the rest of OTel HTTP, database, and messaging conventions. They live next to your existing service telemetry, in the same OTLP pipeline.
When your agent framework — LangChain, CrewAI, AutoGen, Pydantic AI, MCP — emits OTel GenAI spans, every OTel-compatible backend can read them. Trefur is built on this standard end to end. That means you can adopt Trefur without re-platforming, and you can fan out to Datadog, Honeycomb, Grafana in parallel.
Why OpenTelemetry GenAI matters.
Five reasons engineering teams pick the open standard.
One shape across every vendor
The semantic conventions normalise OpenAI, Anthropic, Bedrock, Azure OpenAI, and Gemini into the same span attributes. Compare them, swap them, fall back between them.
Instrument once, capture everywhere
Emit OTel GenAI from your framework callbacks and any OTel backend can read it. No proprietary tracing dialect to support.
Adopt without re-platforming
Your existing OpenTelemetry pipeline reads it natively. Add Trefur as a second exporter; keep Datadog, Honeycomb, Grafana alongside.
Survive the framework churn
Your agent framework will change. Your model vendor will change. OTel GenAI is the layer that stays — your dashboards and alerts keep working.
Open ecosystem, no lock-in
Maintained by the OpenTelemetry community. Trefur consumes it as-is — no proprietary attribute mappings, no shimmed compatibility layer.
How Trefur consumes OpenTelemetry GenAI.
Six places where the OTel GenAI standard shows up in Trefur.
Native OTel GenAI consumption
Trefur reads OTLP/HTTP and OTLP/gRPC on /v1/traces, /v1/metrics, /v1/logs. Standard endpoints.
Read moreSelf-hostable collector
Single Go binary. Reads OTel GenAI on standard endpoints, batches, redacts, fans out.
Read moreVendor-neutral trace shape
Same span attributes for OpenAI, Anthropic, Bedrock, Azure OpenAI, Gemini.
Read moreLangChain emits OTel GenAI
Auto-instrumentation maps LangChain callbacks to OTel GenAI spans automatically.
Read moreMCP emits OTel GenAI
Every MCP server, client, tool call, and resource fetch as OTel spans.
Read moreIngest docs
How the telemetry reaches Trefur — endpoints, headers, authentication, batching.
Read moreSend OTel GenAI to Trefur.
Point your existing OpenTelemetry SDK at the Trefur OTLP/HTTP endpoint. Set the auth header. Spans flow.
export OTEL_EXPORTER_OTLP_ENDPOINT="https://api.trefur.com" export OTEL_EXPORTER_OTLP_PROTOCOL="http/protobuf" export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer trf_obs_..." # Now any OTel-instrumented agent ships to Trefur. # /v1/traces /v1/metrics /v1/logs all read OTel GenAI natively.
Ship OTel GenAI to Trefur today.
Free tier. No card required. First trace in under five minutes.