Exporting and Versioning Flows

Where we left off

flow.json has been sitting in git alongside every lesson's lesson.py since Lesson 2, git add-able, git diff-able, technically a real file like any other. This lesson is about the "technically," what a plain text diff of flow.json actually tells a reviewer (not much), and the two lfx commands worth running on one before it's committed.

Why flow.json doesn't code-review like Python

Open any flow.json in this course and diff it against a version with one field changed, the diff is enormous: every node carries its absolute canvas position, every field its full schema definition restated, a one-line change in intent (a different system prompt) can touch hundreds of lines of JSON. A reviewer reading that diff can't easily tell "what changed" from "what didn't," the way they could with a one-line Python diff. flow.json is worth versioning (it's the actual, reproducible artifact, better than nothing), it's just not a substitute for the review quality plain code gets, that gap is a real part of the prototype-to-production trade-off this course keeps coming back to.

Do this yourself

uv run lfx validate lessons/langflow/01_beginner/02_first_flow/flow.json

Run this yourself. Every flow.json shipped by this course so far fails it, missing a top-level id field, because graph.dump() (what every earlier lesson used to produce its flow.json) doesn't set one, only the server does, when you save or upload a flow through it. This isn't a bug in those lessons, run_flow_from_json and the REST API's upload endpoint both work fine without it, lfx validate is simply a stricter check than either of those requires.

The code, piece by piece

run_lfx("validate", str(SOURCE_FLOW))

lfx validate checks a flow's structural shape, the kind of thing worth running in CI on every flow.json a pull request touches, before anyone even opens it.

flow_data["id"] = str(uuid.uuid4())

The actual fix: a real flow gets its id from the server the moment it's saved or uploaded there, this lesson generates one directly to show validate passing once it's present.

run_lfx("upgrade", str(FLOW_PATH))

A different check: whether each component in the flow still matches what your installed Langflow version expects. Run this after upgrading the langflow/lfx package itself, a component marked BLOCKED here is worth investigating before you trust that flow still behaves the way it did on the older version.