Tutorial: Extract URLs

A practical custom node: paste messy agent output in, get a clean URL list out. Matches the extract-urls demo under examples/nodes/ in the threadle repo. (Smaller starter: import examples/nodes/hello-world first. See Custom nodes. For JSON extraction from agent replies, prefer the stdlib extract-json node.)

Prerequisites: threadle running, and five minutes. For the full descriptor reference see Custom nodes.

1. Create the folder

Settings → custom nodes → ＋ Create node. Name it extract-urls.

Or by hand:

mkdir -p ~/.config/threadle/nodes/extract-urls

2. Descriptor

{
  "id": "extract-urls",
  "label": "Extract URLs",
  "glyph": "⛓",
  "description": "pulls every http(s) URL out of the input, one per line, deduplicated",
  "entry": "node.ts",
  "input": "text",
  "output": "text"
}

3. Logic

export default class ExtractUrls {
  async run(input: string): Promise<string> {
    const urls = input.match(/https?:\/\/[^\s)\]}>"']+/g) ?? [];
    return [...new Set(urls)].join("\n");
  }
}

No threadle imports. Return a string. That becomes the outbound text lane.

4. Try it

Reload threadle (or wait for the nodes watcher).
Palette → nodes → custom → drag Extract URLs.
Wire prompt → Extract URLs → output.
Put this in the prompt:

See https://example.com/a and also https://example.com/a again —
plus https://threadle.sh/docs.

Press ▶ Run. The output node should show:

https://example.com/a
https://threadle.sh/docs

Why this pattern

Piece	Lesson
One text in / one text out	Default contract. No ports or params yet
Pure function	Easy to test mentally. No filesystem, no network
Dedup with `Set`	Nodes are great for boring munging agents forget

Next steps: wrap lines with a param (prefix), or fan into named ports. Or jump to a command node that shells out to sort -u.

Custom nodes
Stdlib pack. extract-json and other utilities
Tutorial: Command node
Tutorial: Typed chain
Params & named ports