Every Node Is One Click From Its Docs

Flow-Like ships 1,600-plus documentation pages — one for every node in the catalog, auto-generated straight from the source. That’s a great library. It’s less great if you can’t find the page for the node under your cursor. This release closes that gap: open a node’s info panel and its documentation is right there.

The docs come to the node

Open the info overlay on any node and, on desktop, you now get an inline preview of that node’s doc page in a side panel; on mobile it renders as a section with an “Open full documentation” button. No searching, no guessing — the docs for this node, immediately.

The interesting part is how the app knows which page to open. There’s no lookup table and no network round-trip to resolve it. The URL is reconstructed from the node itself — its name and category — using the exact same slugify algorithm the docs generator uses to name the pages. Get the node, get the URL:

export function buildNodeDocsUrl(node: INode): string {
  const docsUrl = safeDocsUrl(node.docs);
  if (docsUrl) return docsUrl;
  const categoryPath = categorySegments(node.category).map(safeSegment).join("/");
  const nodeSlug = safeSegment(node.name);
  return `${DOCS_BASE_URL}/nodes/${categoryPath}/${nodeSlug}/`;
}

The safeSegment function is a character-for-character port of the Rust safe_segment in the docs generator. Because both sides slug identically, a node named agent_invoke in category AI/Agents always resolves to docs.flow-like.com/nodes/ai/agents/agent-invoke/ — and that page is guaranteed to exist, because the generator built it from the same catalog. Every one of the 1,600 pages is reachable this way, for free.

This replaced an older, buggy helper that turned node.name into title-cased words with spaces, producing URLs like .../Http Set Bearer Auth that matched nothing. The new path is boring in the best way: deterministic, tested, and correct.

Trust, but verify the URL

A node can carry its own docs field pointing at a custom page. That’s a small trust boundary, so the resolver only accepts http: and https: URLs — anything else (a stray javascript:, say) is ignored and the safe, generated slug is used instead:

function safeDocsUrl(value?: string | null): string | null {
  const docsUrl = value?.trim();
  if (!docsUrl) return null;
  try {
    const url = new URL(docsUrl);
    return url.protocol === "http:" || url.protocol === "https:" ? url.toString() : null;
  } catch { return null; }
}

And the embedded preview iframe was tightened to sandbox="allow-scripts" — dropping same-origin access — so even a compromised or spoofed docs page can’t reach into the app’s WebView.

Faster search on the docs site

While we were in there, the documentation site’s search got a tune-up. It’s a custom Pagefind-backed component that now prewarms on idle instead of waiting until you open it: when the browser is quiet, it quietly loads the index so that the first ⌘/Ctrl+K opens instantly. It also picked up cleaner sub-results and trailing-slash-normalized links along the way.

None of this is a from-scratch docs rewrite — the pages and the generator were already there. What changed is the connective tissue: the app now links precisely into the docs, and the docs are quicker to search. Small pieces, but they’re the ones you hit every day.