Interactive block-based pipeline builder in the docs (drag components → runnable add_pipe / config.cfg)

[rfugal]

The idea

An interactive, block-based pipeline builder embedded directly in the docs. You drag component blocks (tok2vec, tagger, parser, ner, entity_ruler, …) into a pipeline and immediately see the corresponding runnable Python (spacy.blank() / nlp.add_pipe(...)) and a matching config.cfg excerpt.

Why

The pipeline model is one of the highest-value things for a new user to internalize, and one of the hardest to get from static text.

It's already built — and live

Rather than describe it, here's a working prototype you can click right now:

In context, in the docs — a real Blockly canvas embedded on each page:

• Tour of the en_core_web_sm pipeline — https://spacy.fugl.dev/usage/spacy-101
• Build a pipeline, get add_pipe Python (primary) — https://spacy.fugl.dev/usage/processing-pipelines
• Assemble a training config.cfg visually — https://spacy.fugl.dev/usage/training
• Single add_pipe call playground — https://spacy.fugl.dev/api/language

Full standalone composer (adds live execution) — https://spacy.fugl.dev/composer

• The same blocks, plus run it: type a sentence and see displaCy output, via an opt-in in-browser Pyodide engine.
• Open in Composer hand-off: edit a pipeline in any docs embed and open that exact workspace in the full composer in one click.
• Syntax-highlighted Python/config panes, light/dark themes, JSON import/export.

The palette for the built-in components is generated from spaCy's own FactoryMeta, so its ordering, requires/assigns, and "untrained component" warnings come from spaCy itself, not a hand-maintained list that will drift.

Scope and risk

• Touches only the docs site (website/) plus a self-contained standalone app.
• Follows the site's existing interactive-widget pattern (same shape as the quickstart widgets): embedded as a single MDX tag, with all per-page config in one presets file, so placing or removing an embed is a one-line change.
• Renders client-side.

Origin note: this came out of a separate project of mine, levelset — a functional toolkit for authoring resource-leveling algorithms by composing constraint and scoring blocks into schedules. I wanted its users to be able to assemble those block pipelines visually with Blockly, and I chose spaCy deliberately as the prototype domain because it is an ideal model of a pipeline: a clean, well-documented, ordered sequence of components, each with structured, machine-readable metadata (requires/assigns, factory definitions). That made it the perfect testbed for the block vocabulary and code generation — and the spaCy version came out clean and self-contained enough that it seemed worth offering back.

The proposal

Does an interactive pipeline builder belong in the spaCy docs? If so, what shape would you want — which pages, how prominent, is Blockly the right foundation? I'd be glad to open a PR against website/; I just wanted to check appetite and direction before putting up a site-architecture change.