Skip to Content
IntegrationsLangfuse

Langfuse

Langfuse  is an open-source LLM observability platform. auxilia ships with a built-in Langfuse callback that traces every agent run — model calls, tool invocations, token counts, latency, and cost — without any code changes.

What you get

  • Traces — per-message traces showing LLM calls, MCP tool calls, latencies, and token usage
  • Cost tracking — token and cost breakdowns per agent, user, or thread
  • Evaluation — score and annotate agent responses in the Langfuse UI
  • Prompt versioning — manage and compare system prompts across agents

Enable it

Set these three environment variables on the backend:

LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_SECRET_KEY=sk-lf-...
LANGFUSE_BASE_URL=https://cloud.langfuse.com  # or your self-hosted URL

On startup, the backend instantiates a shared Langfuse client and attaches a CallbackHandler to every LangGraph invocation. If any of the three variables is missing, the integration is silently disabled.

See app/integrations/langfuse/callback.py for the wiring.

Self-hosted Langfuse

Langfuse is itself open source. You can run it alongside auxilia for full data ownership — follow the Langfuse self-hosting guide  and point LANGFUSE_BASE_URL at your instance.

OverviewSlack