Agent Plugin
Hermes Langfuse Observability
Exports Hermes conversations, model calls, and tool activity to Langfuse while failing open when telemetry is unavailable.
What is Hermes Langfuse Observability?
Langfuse Observability is Hermes Agent's opt-in tracing plugin. It instruments a session across conversation, model, and tool boundaries, then sends structured traces to a configured Langfuse project. Sampling, environment, release, payload size, and endpoint settings can be tuned, while missing credentials or SDK failures do not block the agent's primary work.
Key capabilities
End-to-end traces
Correlates session activity across model calls and tool execution.
Configurable telemetry
Supports sampling, release, environment, endpoint, and payload-size controls.
Fail-open operation
Observability failures do not interrupt the Hermes task itself.
Use cases
Agent debugging
Inspect where a task spent time and how tool calls affected the result.
Production monitoring
Track latency, usage, and failures across deployed Hermes sessions.
Prompt evaluation
Compare trace behavior across releases or environments.
Included components
Hooks
Langfuse Tracing
Emits conversation, LLM-call, and tool-call spans to Langfuse.
Installation scope
- Hermes Agent
- The langfuse Python package
- HERMES_LANGFUSE_PUBLIC_KEY and HERMES_LANGFUSE_SECRET_KEY
- A Langfuse project or compatible self-hosted endpoint