Setup and First Trace

What we're building

A single function decorated with @traceable, so that calling it once sends a record of the call to LangSmith. This is the smallest possible trace, one function, one input, one output, and it's the foundation every later lesson builds on.

Why a separate service

LangChain and LangGraph, the last two courses, are libraries: code that runs on your machine. LangSmith is a hosted service: a place outside your machine that stores what your code did, so you can look at it after the fact from a browser. You already have a GOOGLE_API_KEY in .env for talking to Gemini; now you need a second, unrelated key, LANGSMITH_API_KEY, for talking to LangSmith. Get one free at smith.langchain.com, then add to .env:

LANGSMITH_API_KEY=your-key-here
LANGSMITH_TRACING=true

LANGSMITH_TRACING=true is the master switch. Without it, @traceable functions run exactly as normal Python, nothing is sent anywhere.

The code, piece by piece

from langsmith import traceable

traceable is a decorator, a function that wraps another function to add behavior around it without changing its body.

@traceable
def summarize(text: str) -> str:
    ...

Putting @traceable directly above a function means: every time summarize(...) is called, first record the arguments it was given, then run the real function body, then record what it returned (or, if it raised an exception, record that instead). All of this happens in the background over the network; the function's own behavior is unchanged.

result = summarize("LangSmith records ...")

An ordinary function call. Nothing about calling summarize looks any different from calling any other function, tracing is invisible from the caller's side.