Langfuse
Open-source LLM observability platform — traces, costs and evaluations; I wire it into LangGraph services.
Langfuse is an open-source platform for LLM engineering: tracing of prompts, model and tool calls, cost and latency tracking, and evaluation scores.
How I use it
- langgraph-langserve-starter — set two keys and a Langfuse callback handler mounted on the FastAPI app traces every invoke and stream of the LangGraph agent.
- Conversational AI Agents — Langfuse/OpenTelemetry tracing alongside automated Evals in my Independent AI & Full-Stack Engineer (Freelance) work.
Why it matters
Langfuse is open source and self-hostable, which makes it a practical default for client projects that cannot send prompts and outputs to a third-party SaaS. Paired with OpenTelemetry, it gives a vendor-neutral path for traces, costs and evaluation scores.
In my starter template it is strictly optional: without keys the service runs untraced, so the template stays zero-config while being production-ready when observability is switched on.
Related
-
AI Agent Products & Remote MCP — production use in my freelance work