Hugging Face Scraper

Models, datasets and Spaces from the Hugging Face Hub, with downloads, likes, licenses, parameter counts and base models, as clean JSON.

Hugging Face is where the open AI ecosystem publishes its work: over a million models, hundreds of thousands of datasets, and the Spaces that demo them. This Actor scrapes AI models, datasets and Spaces from Hugging Face: downloads, likes, trending score, license, task, library, parameter count, base models, languages and Space status. Paste any Hub listing with its filters, an organization page, or a single model, dataset or Space, or just search by keyword. Each repo comes back with its last-30-day and all-time downloads, likes and trending score, creation and update dates, gating, the papers it cites, and for Spaces the SDK, hardware and whether the app is running. There is no login, no API key, and no proxy to configure. Built for AI market research, model and vendor tracking, open-source analytics, investors watching what developers adopt, and teams building model catalogs.

⚡ Quick start

{
  "startUrls": [
    {
      "url": "https://huggingface.co/models?pipeline_tag=text-generation&sort=trending&search=llama"
    }
  ],
  "searchTerms": ["whisper"],
  "maxItems": 5
}

🧩 Input

Field	Type	Required	Default	Description
startUrls	array	no	—	huggingface.co pages: a model, dataset or Space listing with its filters (/models?…, /datasets?…, /spaces?…), an organization or user (/meta-llama), or a model, dataset or Space page.
searchTerms	array	no	—	Keywords to search the Hub for, one per line, for example whisper or stable diffusion.
repoType	string	no	model	What search terms look for: model, dataset or space.
sort	string	no	trending	Order of search results: trending, downloads, likes, created or modified. Spaces have no downloads, so for Spaces downloads sorts by likes.
task	string	no	—	Only return models or datasets for one task, such as text-generation, automatic-speech-recognition or text-to-image. Ignored for Spaces.
maxItems	integer	no	—	Maximum number of repos per start URL or search term. The input form starts at 5; leave it empty for every matching repo.
proxyConfiguration	object	no	{ "useApifyProxy": false }	Proxy settings. The Hugging Face Hub needs no proxy.

Add URLs, search terms, or both. The easiest way to build a query is to filter on huggingface.co/models (task, library, language, license, size, sort, search) and paste the address bar: the Actor keeps every one of those filters, and the same goes for the datasets and Spaces pages. An organization or user page such as https://huggingface.co/google returns that account's models, most downloaded first. The search options (Search in, Sort by, Task) apply to search terms only.

📤 Output

Each model, dataset or Space is one dataset row, marked by itemType. Every row has the same fields; a field that does not apply to that kind of repo (downloads on a Space, parameters on a dataset) is null.

Fields

Field	Type	Description
itemType	string	model, dataset or space.
id	string	Repo ID, owner/name.
author	string	Owner: organization or user.
name	string	Repo name without the owner.
url	string	The repo's page on huggingface.co.
title	string	Display title of a dataset or Space. null for models.
description	string	Dataset summary as plain text, or a Space's short description. null for models.
pipelineTag	string	A model's task, such as text-generation.
taskCategories	array	A dataset's task categories, such as question-answering.
library	string	A model's library, such as transformers, diffusers or gguf.
tags	array	Every tag on the repo.
license	string	License ID, such as apache-2.0 or llama3.1.
languages	array	Languages declared on the repo card.
downloads	number	Downloads over the last 30 days (models and datasets).
downloadsAllTime	number	Downloads since the repo was created (models and datasets).
likes	number	Likes.
trendingScore	number	Hugging Face's trending score: recent likes and activity.
createdAt	string	When the repo was created (ISO 8601).
lastModified	string	When the repo was last updated (ISO 8601).
gated	boolean	Whether you must accept terms or be approved to download the files (models and datasets).
gatedMode	string	auto (accept terms) or manual (owner approves) when gated.
parameterCount	number	Total model parameters, from the safetensors or GGUF weights. null when the weights don't publish it.
baseModels	array	The model(s) this model was built from.
baseModelRelation	string	How it was built from them: finetune, quantized, adapter or merge.
datasets	array	Datasets a model was trained on (as declared on its card), or datasets a Space uses.
models	array	Models a Space uses.
arxivIds	array	arXiv IDs of the papers the repo cites.
sizeBytes	number	Size of a dataset's main files in bytes.
sdk	string	A Space's SDK: gradio, streamlit, docker or static.
runtimeStage	string	Whether a Space is up: RUNNING, SLEEPING, BUILDING, RUNTIME_ERROR and so on.
hardware	string	Hardware a Space runs on, such as cpu-basic or zero-a10g.
emoji	string	A Space's emoji.
appUrl	string	The Space's app on its own domain (*.hf.space).

💡 Use cases

• AI market research — track which models, tasks and licenses are gaining downloads and likes, week over week.
• Vendor and competitor tracking — monitor every release from an organization such as meta-llama, google or mistralai, with parameter counts and licenses.
• Model lineage — map fine-tunes and quantizations back to their base models to measure a model family's reach.
• Dataset discovery — build catalogs of datasets by task, language, license and size.
• Investment and partnership research — spot fast-rising authors and Spaces by trending score.

❓ FAQ

• Do I need a Hugging Face account or API token? No. The Actor reads the same public pages anyone can browse without logging in.
• Do I need a proxy? No. It runs without one by default.
• What do the download numbers mean? downloads counts the last 30 days, the number Hugging Face shows on a model page; downloadsAllTime counts every download since the repo was created.
• Why is parameterCount empty for some models? It is read from the published weight files. Models without safetensors or GGUF weights (older PyTorch-only checkpoints, adapters, API-only models) don't publish it.
• Can I get gated models like Llama? Yes, their public details: downloads, likes, license and so on. gated tells you that downloading the weights needs approval; the Actor never downloads files.
• How many repos can I get? As many as the listing has. Leave maxItems empty to page through every match; each page holds 100 repos. Hugging Face limits how fast anonymous clients can browse, so very large runs pause briefly when they hit that limit and then continue.

🔗 More scrapers by Anyx

• Hacker News Scraper
• Web Content Scraper
• Reddit Scraper
• Google Search Scraper

🤝 Anyx Solutions

We build custom scrapers and data-extraction pipelines.

• Email: tantosthor@gmail.com

🖼 Image credit

Image credit: huggingface.co