LangFuse — Open Source LLM Observability & Tracing

What is LangFuse?

LangFuse is an open-source observability platform for LLM applications. It provides tracing, prompt management, evaluation, and cost analytics — helping teams debug, improve, and monitor their AI features in production.

Answer-Ready: LangFuse is an open-source LLM observability platform providing tracing, prompt management, evaluation pipelines, and cost analytics for production AI applications.

Core Features

1. Distributed Tracing

Trace complex chains and agent workflows:

from langfuse.decorators import observe

@observe()
def my_agent(query: str):
    context = retrieve_docs(query)
    return generate_response(query, context)

@observe()
def retrieve_docs(query: str):
    # Automatically nested as child span
    return vector_db.search(query)

@observe()
def generate_response(query: str, context: str):
    return openai.chat.completions.create(...)

2. Framework Integrations

# OpenAI SDK (drop-in)
from langfuse.openai import openai
# All calls automatically traced

# LangChain
from langfuse.callback import CallbackHandler
handler = CallbackHandler()
chain.invoke({"input": "..."}, config={"callbacks": [handler]})

# LlamaIndex
from llama_index.core import Settings
Settings.callback_manager.add_handler(langfuse_handler)

3. Prompt Management

Version and deploy prompts from the LangFuse UI:

prompt = langfuse.get_prompt("customer-support-v2")
compiled = prompt.compile(customer_name="Alice")

4. Evaluation Pipelines

Score traces manually or with LLM-as-judge:

langfuse.score(
    trace_id="trace-123",
    name="helpfulness",
    value=0.9,
    comment="Accurate and complete",
)

5. Cost Dashboard

Automatic cost calculation per model, per user, per feature.

Self-Hosting

docker compose up -d  # PostgreSQL + LangFuse server

Or use the managed cloud at cloud.langfuse.com.

FAQ

Q: How does it compare to LangSmith? A: LangFuse is open-source and self-hostable. LangSmith is LangChain-specific and proprietary.

Q: Does it work without LangChain? A: Yes, framework-agnostic. Works with any Python or JS/TS app.

Q: Production overhead? A: Async by default — traces are batched and sent in background with < 1ms overhead.