lfx Headless Execution

Where we left off

Every earlier lesson imported Langflow components from inside a Python script (from lfx.components... import ...). lfx is also its own command-line tool, installed alongside langflow in this repo but usable entirely on its own: point it at a flow.json and an input, get an answer back, no Python script required at all.

Why this matters

A CI pipeline that just needs to smoke-test a flow, or a cron job that runs one on a schedule, doesn't need the full langflow package, a running server, or even a .py file. lfx run is the smallest possible way to execute a flow: one shell command, one process, exits when it's done.

Do this yourself

uv run lfx run lessons/langflow/01_beginner/02_first_flow/flow.json \
  "Say hello in exactly three words." -f text

Run this directly in your terminal, no lesson.py involved. -f text prints just the answer, drop it (or try -f json) to see the full structured result Langflow builds internally, the same shape run_flow_from_json returns as a Python object.

The code, piece by piece

subprocess.run(
    [sys.executable, "-m", "lfx", "run", str(FLOW_PATH), input_value, "-f", "text"],
    capture_output=True,
    text=True,
    check=True,
)

lesson.py shells out to the exact same command from "Do this yourself", via subprocess, so this repo's usual uv run python lesson.py convention keeps working. In a real CI pipeline you'd run the lfx run ... line directly, this wrapper exists only so this course's own pattern (one runnable lesson.py per lesson) holds here too.