Get Started

This guide walks you from a blank Muster install to a trace visible in the UI. You'll need a running Muster instance — either the public hosted version or your own self-hosted deployment (see Self-hosting).

Step 1 — Create a project and grab API keys

1. Log in to Muster (the URL is whatever your operator gave you, or http://localhost:3000 for a local install).
2. Create an organization, then create a project inside it.
3. Open Settings → API Keys in the project sidebar.
4. Click Create new API key.
5. Copy both keys somewhere safe:
   • Public key (pk-lf-...) — used in client SDKs
   • Secret key (sk-lf-...) — used in server SDKs, never expose to browsers

The secret key is shown only once. If you lose it, rotate the key and create a new one.

Step 2 — Install the SDK

Pick the language you're instrumenting.

Python:

pip install langfuse

TypeScript / JavaScript:

npm install langfuse

The SDKs are the same packages used by upstream Langfuse — Muster speaks the identical ingestion API.

Step 3 — Send a trace

Python:

from langfuse import Langfuse

langfuse = Langfuse(
    public_key="pk-lf-...",
    secret_key="sk-lf-...",
    host="http://localhost:3000",  # your Muster URL
)

trace = langfuse.trace(
    name="user_question",
    user_id="user_123",
    metadata={"environment": "production"},
)

trace.generation(
    name="llm_response",
    model="gpt-4",
    input="What is AI?",
    output="AI stands for Artificial Intelligence...",
)

langfuse.flush()

TypeScript:

import { Langfuse } from "langfuse";

const langfuse = new Langfuse({
  publicKey: "pk-lf-...",
  secretKey: "sk-lf-...",
  baseUrl: "http://localhost:3000",
});

const trace = langfuse.trace({
  name: "user_question",
  userId: "user_123",
  metadata: { environment: "production" },
});

trace.generation({
  name: "llm_response",
  model: "gpt-4",
  input: "What is AI?",
  output: "AI stands for Artificial Intelligence...",
});

await langfuse.flush();

flush() blocks until queued events are delivered. Don't skip it in short-lived scripts — the SDK batches by default and processes can exit before the buffer sends.

Step 4 — View the trace

Open the Traces page in the project sidebar. Your trace should appear within a few seconds. Click into it to see the nested observation, input, output, and timing.

If nothing appears:

• Check the SDK logs for HTTP errors (4xx usually means the API key is wrong; 5xx usually means the server-side ingestion worker isn't running).
• Confirm the host / baseUrl matches your Muster URL exactly, including protocol.
• Self-hosters: confirm the muster-worker container is running. Traces are ingested asynchronously — if the worker is down, traces never arrive.

What's next

• Read Concepts to understand the data model: trace, observation, score, session, dataset.
• Once you have traces flowing, set up Agent Inventory to register the agents producing those traces.
• If you also want Muster to find agents you didn't instrument, configure the Discovery Engine.