Capability · Framework — observability

Langfuse

Langfuse is an open-source alternative to LangSmith. It provides distributed tracing for LLM apps (OpenAI, Anthropic, LangChain, LlamaIndex, Haystack, DSPy, Vercel AI SDK, and more via OTel), prompt management with versioning and A/B testing, datasets, online and offline evaluations, and dashboards for cost/latency/quality. You can self-host with Docker or use Langfuse Cloud (EU/US).

Framework facts

Category: observability
Language: Python / TypeScript
License: MIT (core) / EE
Repository: https://github.com/langfuse/langfuse

Install

pip install langfuse
# or
npm install langfuse

Quickstart

from langfuse import Langfuse, observe

langfuse = Langfuse(
    public_key='pk-lf-...',
    secret_key='sk-lf-...',
    host='https://cloud.langfuse.com'
)

@observe()
def answer(q: str) -> str:
    return f'echo: {q}'

answer('hello')

Alternatives

LangSmith — LangChain's commercial alternative
Arize Phoenix — Apache-licensed, OTel-native
Helicone — proxy-based observability
Traceloop — OTel-based LLM tracing

Frequently asked questions

Is the self-hosted Langfuse fully featured?

The core (tracing, datasets, prompts, evals) is MIT-licensed and self-hostable. A small set of advanced features (SSO/SAML, advanced RBAC, analytics) are enterprise-only. Most teams run the self-hosted core in production successfully.

Langfuse or LangSmith?

Langfuse if you want open-source, self-hosted, or framework-agnostic LLM observability with strong EU/US cloud options. LangSmith if you're all-in on LangChain/LangGraph and want a tightly integrated commercial experience. Both are excellent.

Sources

Langfuse — docs — accessed 2026-04-20
Langfuse — GitHub — accessed 2026-04-20