· Doc version 1.0

Langfuse vs LangSmith: choosing an LLM observability stack

Both trace, evaluate, and monitor LLM agents — but they differ on hosting, openness, and ecosystem. Here's how to choose.

Related concepts

Once an agent does more than one model call, logs stop being enough: you need to see the trace — every model and tool step, with inputs, outputs, latency, and cost. Two tools dominate that space for AI agents: Langfuse and LangSmith. They overlap a lot, so the decision usually comes down to a few axes.

At a glance

LangfuseLangSmith
SourceOpen source (MIT)Proprietary (SaaS)
HostingCloud or self-hostHosted (cloud / enterprise)
EcosystemFramework-agnosticLangChain-native, also standalone
CoreTracing, evals, prompt mgmt, analyticsTracing, evals, prompt mgmt, monitoring
PricingOpen source + free cloud tierFreemium (free tier + paid)

Where Langfuse shines

Langfuse is open source, so you can self-host and keep prompts and user data inside your own infrastructure — often the deciding factor in regulated or privacy-sensitive settings. It’s deliberately framework-agnostic: the SDK and OpenTelemetry support drop into any stack, whether or not you use LangChain.

Where LangSmith shines

LangSmith is the most frictionless option if you’re already building on LangChain or LangGraph — tracing lights up with almost no wiring. As a managed product it leans into a polished, batteries-included experience for datasets, evaluations, and production monitoring.

How to choose

  • Want to self-host or stay framework-agnostic? → Langfuse.
  • Already on LangChain/LangGraph and want zero-setup tracing? → LangSmith.
  • Need open source / data residency control? → Langfuse.
  • Want a fully managed, polished eval suite? → LangSmith.

Both are strong; many teams start with whichever matches their framework and hosting constraints, since the day-to-day tracing experience is comparable.

Related tools