Logfire Observability

This example instruments an HNO agent with OpenTelemetry and forwards spans to Logfire. It is useful when you want to correlate reasoning content, token usage, and tool activity inside your observability platform.

Requirements

• Go 1.21+
• OpenAI API key (or another reasoning-capable provider)
• Logfire write token (LOGFIRE_WRITE_TOKEN)
• Optional: custom OTLP endpoint (LOGFIRE_OTLP_ENDPOINT defaults to logfire-eu.pydantic.dev)

Run

export OPENAI_API_KEY=sk-your-key
export LOGFIRE_WRITE_TOKEN=lf_your_token
go run -tags logfire cmd/examples/logfire_observability/main.go

Note
The example is guarded by the logfire build tag to avoid pulling OpenTelemetry dependencies during normal builds.

What the Example Does

1. Configures an OTLP/HTTP exporter with TLS and the Logfire write token.
2. Runs a reasoning-capable agent (OpenAI o1 preview by default).
3. Records span attributes for runtime, loop count, and token usage.
4. Emits a reasoning.complete span event containing the reasoning snippet (and optional redacted content).

In Logfire you will see spans similar to:

• agent.run
  • Attributes: agent.model, agent.provider, agent.duration_ms, agent.usage.*
  • Events: reasoning.complete with reasoning.content, reasoning.token_count

Related Docs

• docs/release/logfire_observability.md – deep dive guide (GitHub).
• website/advanced/observability.md – overview of the observability stack, including the SSE event stream.