Langfuse Cheatsheet

Q: How do I set up Langfuse?

Self-host: `git clone https://github.com/langfuse/langfuse.git && cd langfuse && docker compose up -d`. Cloud: sign up at cloud.langfuse.com. Set env vars: `LANGFUSE_PUBLIC_KEY`, `LANGFUSE_SECRET_KEY`, `LANGFUSE_HOST`.

Q: What is a trace vs a span?

A trace represents one LLM operation (e.g., a chat completion call). A span is a sub-step within a trace (e.g., vector DB query, tool call). Use traces for top-level operations and spans for nested sub-operations. Generations are spans specifically for LLM calls with token tracking.

Q: How does Langfuse help with prompt management?

Langfuse provides version-controlled prompt templates. Create prompts in the dashboard, then fetch them via SDK (`lf.get_prompt('name')`). Prompts support variables, version labels, and automatic rollback. Changes can be tested without redeploying.

Langfuse is an open-source LLM observability platform offering trace/span tracking, token usage statistics, scoring, prompt version management, and dataset management. Supports self-hosting and cloud with deep LangChain and OpenAI integration.

Updated: 2026-07-20·14 commands

Quick Start

``bash pip install langfuse export LANGFUSE_PUBLIC_KEY=pk-lf-xxx export LANGFUSE_SECRET_KEY=sk-lf-xxx export LANGFUSE_HOST=https://cloud.langfuse.com``

Core Concepts

Traces represent end-to-end LLM operations. Spans break traces into sub-steps. Generations track LLM calls with prompt/completion tokens. Scores enable human or automated evaluation. Datasets hold test cases for eval runs.

Startup & Modes(12)

Command	Level	Example
`langfuse --help` Show Langfuse CLI help	Basic	`langfuse --help`
`pip install langfuse` Install Langfuse Python SDK	Basic	`pip install langfuse`
`git clone langfuse` Clone Langfuse source	Basic	`git clone https://github.com/langfuse/langfuse.git`
`export LANGFUSE_PUBLIC_KEY` Set Langfuse public key	Basic	`export LANGFUSE_PUBLIC_KEY=pk-lf-xxx`
`export LANGFUSE_SECRET_KEY` Set Langfuse secret key	Basic	`export LANGFUSE_SECRET_KEY=sk-lf-xxx`
`export LANGFUSE_HOST` Set Langfuse server URL	Basic	`export LANGFUSE_HOST=http://localhost:3000`
`langfuse trace` Create a Langfuse trace	Intermediate	`python -c 'from langfuse import Langfuse; lf = Langfuse(); trace = lf.trace(name="my-llm-call"); trace.update(output="result")'`
`langfuse span` Create a span (sub-step of a trace)	Intermediate	`python -c 'from langfuse import Langfuse; lf = Langfuse(); trace = lf.trace(name="pipeline"); span = trace.span(name="llm-call"); span.end()'`
`langfuse generation` Track LLM generation (with token counting)	Intermediate	`python -c 'from langfuse import Langfuse; lf = Langfuse(); lf.generation(name="chat", model="gpt-4", input="Hello", output="Hi", usage={"promptTokens":10,"completionTokens":5})'`
`langfuse score` Add a score to a trace	Expert	`python -c 'from langfuse import Langfuse; lf = Langfuse(); lf.score(name="accuracy", value=0.95, trace_id="xxx")'`
`langfuse prompt` Get managed prompt template from Langfuse	Intermediate	`python -c 'from langfuse import Langfuse; lf = Langfuse(); prompt = lf.get_prompt("my-prompt"); print(prompt.compile(input="value"))'`
`langfuse dataset` Create or get an evaluation dataset	Expert	`python -c 'from langfuse import Langfuse; lf = Langfuse(); dataset = lf.get_dataset("eval-data"); print(len(dataset.items))'`

ai-server(2)

Command	Level	Example
`langfuse server` Start Langfuse local server (Docker)	Intermediate	`langfuse server`
`docker compose langfuse` Start Langfuse with Docker Compose	Intermediate	`docker compose -f docker-compose.yml up -d`

FAQ

How do I set up Langfuse?

What is a trace vs a span?

How does Langfuse help with prompt management?

This cheatsheet is compiled from official tool documentation. Last updated: 2026-07-20.