Getting Started with Langfuse for LLM Observability: A Step-by-Step Tutorial

Getting Started with Langfuse for LLM Observability

If you are building LLM-powered applications, understanding what your models are actually doing in production is critical. Langfuse provides a complete observability and evaluation platform that gives you deep visibility into every LLM call, agent step, and RAG retrieval — all from a clean, self-hosted dashboard. This tutorial walks you through setting up Langfuse and connecting your first application.

Prerequisites

• Docker installed on your server or development machine
• A Python or Node.js application using an LLM (OpenAI, Anthropic, open-source models)
• Basic familiarity with environment variables

Step 1: Deploy Langfuse with Docker Compose

Create a docker-compose.yml file:

services:
  langfuse:
    image: langfuse/langfuse:latest
    restart: unless-stopped
    ports:
      - "3000:3000"
    environment:
      - NEXTAUTH_SECRET=your-secret-here
      - SALT=your-salt-here
      - DATABASE_URL=postgresql://postgres:***@db:5432/langfuse
      - TELEMETRY_ENABLED=false
    depends_on:
      db:
        condition: service_healthy

  db:
    image: postgres:15
    restart: unless-stopped
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
      POSTGRES_DB: langfuse
    volumes:
      - postgres_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 5s
      timeout: 5s
      retries: 5

volumes:
  postgres_data:

Run: docker compose up -d. Langfuse will be available at http://localhost:3000.

Step 2: Configure Your First Project

• Open the Langfuse dashboard and create an account (first user is admin)
• Create a new project (e.g., "My LLM App")
• Navigate to Settings → API Keys and generate a public/secret key pair

Step 3: Instrument Your Python Application

Install the Langfuse Python SDK:

pip install langfuse openai

Add tracing to your OpenAI calls:

from langfuse import Langfuse
from openai import OpenAI

langfuse = Langfuse(
    public_key="pk-...",
    secret_key="sk-...",
    host="http://localhost:3000"
)

client = OpenAI()

# Create a trace for your LLM call
trace = langfuse.trace(name="my-rag-query")

# Log a generation
generation = trace.generation(
    name="openai-response",
    model="gpt-4o",
    input="What is Langfuse?",
    output="Langfuse is an open-source LLM engineering platform..."
)

Step 4: Explore the Dashboard

Back in the Langfuse UI, you will see your first trace appear in the Traces tab. From here you can:

• Filter by model, user, or date range
• Inspect individual trace details including latency and token costs
• Create custom dashboards for your key metrics
• Set up automated evaluations to score responses

Conclusion

You now have a fully self-hosted LLM observability stack running in minutes. Langfuse gives you the same level of insight into your AI applications that traditional APM tools provide for web services — tracing, metrics, evaluation, and debugging all in one place. As your application grows, explore Langfuse's advanced features: dataset management for regression testing, prompt versioning for controlled rollouts, and annotation queues for human-in-the-loop evaluation.