Hugging Face Hub Scraper

Hugging Face Hub scraper for models, datasets and Spaces as clean JSON, no login or API key. Search the Hugging Face Hub, filter by task / library / license / author, sort by downloads, likes or trending score, or look up exact repo IDs. It uses the Hub's public JSON API, so runs are fast and you pay per result.

What you can do with it

• Model & dataset discovery — search: "llama", filter: ["text-generation", "gguf"], sort by downloads to find what the community actually uses.
• Trend tracking — schedule sort: trendingScore daily and diff new entries to spot rising models and Spaces before they hit the news.
• Competitor / org monitoring — author: "meta-llama" lists every repo an org publishes with downloads, likes and last-modified dates.
• Portfolio watch — pass ids for the exact repos you depend on and track downloads, likes, gating and license over time.
• Automation — Schedule it on Apify and pipe results to Google Sheets, Slack, n8n or Make via the Apify integration.
• Model selection for a project — filter: ["license:apache-2.0", "text-generation"] sorted by likes gives a shortlist of permissively licensed models with their base model and library.

Output example

{
  "type": "model",
  "id": "meta-llama/Llama-3.1-8B-Instruct",
  "author": "meta-llama",
  "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct",
  "downloads": 4501381,
  "likes": 8729,
  "trending_score": null,
  "pipeline_tag": "text-generation",
  "library_name": "transformers",
  "tags": ["transformers", "safetensors", "llama", "text-generation", "license:llama3.1", "…"],
  "created_at": "2024-07-18T08:56:00.000Z",
  "last_modified": "2024-09-25T17:00:57.000Z",
  "gated": "manual",
  "private": false,
  "sha": "0e9e39f249a16976918f6564b8830bc894c89659",
  "sdk": null,
  "card_data": { "license": "llama3.1", "language": null, "base_model": "meta-llama/Llama-3.1-8B" }
}

type is model, dataset or space. sdk (gradio, streamlit, docker …) is set for Spaces. gated is false, "auto" or "manual".

Input

Field	Description
mode	models (default), datasets, spaces
search	Keyword matched against repo names.
author	User or org name, e.g. google.
filter	List of Hub tags, all must match: task (text-generation), library (pytorch, gguf), license:mit, language (en) …
sort	downloads (default), likes, lastModified, trendingScore. Always descending.
ids	Exact repo IDs (or Hub URLs). When set, search / author / filter / sort are ignored. Unknown or private IDs are skipped with a warning.
maxItems	Stop after N rows. Default 100.

Pricing

Pay per item: one event per dataset row (model, dataset or Space). Nothing is charged for rows that are never produced.

Notes

• downloads is the Hub's rolling 30-day count; trending_score is only filled when the Hub reports one.
• card_data comes from the repo card when listed, otherwise from license: / base_model: tags; language only from the card.
• Anonymous requests see public repos only.

FAQ

Do I need an API key or account?

No. Anonymous access to the public Hub API is enough. The Hub applies its usual anonymous rate limits; normal runs of a few hundred rows stay well inside them.

How much does it cost?

$0.002 per result and nothing else: with pay-per-event pricing you are not billed for Apify compute. 100 results cost $0.20, and the Apify free plan's $5 monthly credit covers about 2,500 results. Rows that are never produced are never charged.

How do I get the data into Google Sheets / my app?

Every run writes to an Apify dataset you can download as CSV, JSON or Excel, or read from the dataset API endpoint. For ongoing feeds, use Apify integrations (Google Sheets, Slack, Zapier, Make, n8n) or call the Actor from your own code with the Apify API.

Is it legal / allowed?

It only reads public data via the platform's public API and respects rate limits. You are responsible for complying with Hugging Face's terms and applicable law.

Why is one of my ids missing from the results?

Anonymous requests can only see public repos. The Hub answers 401 for both private and non-existent repos, so the Actor skips those IDs with a warning in the log and charges nothing for them. Gated repos (gated: "manual" or "auto") are public metadata and are returned normally.

Related Actors

• npm / PyPI Package Checker
• DEV Community Articles Scraper
• Wikipedia Scraper