spruce_the_data: messy HTML and text to clean JSON

Scraped pages are messy: leftover HTML, menus, cookie banners, prices written five different ways. spruce_the_data reads that mess and returns tidy JSON in exactly the shape you ask for. You describe the fields you want with a JSON Schema, and every result matches it.

• Your schema, your fields. Products, job ads, events, contacts, listings: anything you can describe as a JSON object.
• Give it links or text. Paste web page URLs and it downloads and cleans each page, or send text and HTML you already have.
• Works with any scraper. Point it at another Actor's dataset and it cleans every item.
• Validated output. Results are checked against your schema. Emails, dates, URLs and phone numbers get a format check too; a nullable field that still fails comes back as null and is listed in droppedFields, so you know which values to double-check.
• Failed items are free. You only pay for items that come back clean.
• Built-in prompt-injection defenses. Instructions hidden in a page ("ignore previous instructions...") are treated as data, which makes them hard to use against you.

Quick start

1. Paste one or more messy texts or HTML snippets into Texts to clean, or web addresses into Web page URLs.
2. Describe the shape you want in Output JSON Schema (or leave it empty for a general title, summary and entities extraction).
3. Click Start. Results appear in the Output tab, one row per input.

Example input

{
  "texts": [
    "<h1> Stellar Gooseneck Kettle 1.0L </h1><span class=\"price\">$49.99</span> In stock"
  ],
  "outputSchema": {
    "type": "object",
    "properties": {
      "name": { "type": "string" },
      "price": { "type": ["number", "null"] },
      "currency": { "type": ["string", "null"], "description": "ISO 4217 code, e.g. USD" },
      "in_stock": { "type": ["boolean", "null"] }
    },
    "required": ["name", "price", "currency", "in_stock"]
  }
}

Example output

{
  "index": 0,
  "status": "success",
  "data": {
    "name": "Stellar Gooseneck Kettle 1.0L",
    "price": 49.99,
    "currency": "USD",
    "in_stock": true
  },
  "droppedFields": []
}

• index is the position of the item in your input, so you can match results back.
• droppedFields lists any fields that failed a format check and were set to null.
• A failed item looks like {"index": 3, "status": "failed", "error": "..."} and is not charged.

Cleaning web pages by URL

Put one address per line in Web page URLs:

{
  "urls": ["https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html"],
  "outputSchema": {
    "type": "object",
    "properties": {
      "title": { "type": "string" },
      "price": { "type": ["number", "null"] },
      "currency": { "type": ["string", "null"], "description": "ISO 4217 code, e.g. GBP" },
      "in_stock": { "type": ["boolean", "null"] }
    },
    "required": ["title", "price", "currency", "in_stock"]
  }
}

Each result row also has a url field, so you can match it to the page.

• Best for single item pages: one product, job ad, listing or profile per URL.
• Plain downloads only. Pages that need JavaScript to show their content, need a login, or block automated visitors won't load. Those items fail with a message and are free; for such sites, run a full scraper first and clean its dataset (below).
• Pages up to 2 MB, 45 seconds to load, public http and https addresses only.

Cleaning another Actor's results

Already have a scraper? Run it first, then:

1. Copy its dataset ID (on the run's Storage tab).
2. Paste it into Source dataset ID here.
3. Set Text field in the source dataset to the field that holds the page text or HTML (for example text, html or description).

Every item in that dataset is cleaned into your schema, up to Max items.

Tips for accurate results

• Add a description to fields. "ISO 4217 code, e.g. USD" or "price in the listing's own currency, as a number" makes a real difference.
• Use nullable types (["number", "null"]) for anything that might be missing, so the model can say "not there" instead of guessing.
• Send raw HTML if you have it. Scripts, styles and layout are stripped automatically, and useful attributes (like a star-rating class) are kept as hints.
• One record per item works best. Long list pages with dozens of entries work, but are slower and count as large items.

Pricing

You pay per successfully cleaned item. Failed items are always free.

• Item cleaned: charged once per item that comes back clean.
• Large item (extra): added when an item's output is large (over about 600 tokens), such as a long list page, because those take several times longer to process.
• Run start: a tiny fee per run, set by Apify.

The current prices are shown on this Actor's Pricing tab. You can cap the cost of any run with Apify's maximum cost per run setting; the Actor stops cleanly when the cap is reached.

Users on Apify's free plan can clean 5 items per run to try it out.

Limits and speed

• About 5 seconds per typical item; long list pages can take 30 to 50 seconds.
• Items are processed one at a time, so large batches take a while. Start them in the background and come back.
• Each input can be up to 300,000 characters of raw HTML or text. After cleanup, the part the model reads is capped at 20,000 characters; longer items fail (free) with a message.
• Ratings shown only as star images with no text or attributes can't be read.
• Link addresses (href) and image URLs inside a page aren't read, only the visible text. If you need one, include it in your text or take it from your scraper's own output.
• The service is occasionally unavailable for maintenance. Busy or unavailable moments are retried automatically; anything that still fails is not charged.

Privacy

Your texts are used only to produce your results. The cleaning service doesn't store or log them and doesn't use them to train models. Your inputs and results stay in your own Apify storage, under your account's retention settings.

Questions or problems?

Open an issue on this Actor's Issues tab with an example input (remove anything private) and what you expected to get back.