Hugging Face Scraper

Search and extract Hugging Face model, dataset, and Space listings, full repo details (files, config, card metadata), and user/organization profiles — no account or API key needed.

Why use this actor

• One actor covers all of Hugging Face's public surface: search models/datasets/Spaces, pull full download-ready details for a specific repo, and look up user or organization profiles
• Full repo detail includes the complete file manifest (siblings), parsed model/dataset card metadata, config, and (for Spaces) the live demo URL
• No account, login, or API key required — this actor talks to Hugging Face's own official public REST API
• Automatic retries on temporary network hiccups, including respecting the platform's own rate-limit headers
• Robust pagination — automatically follows the platform's Link-header pagination up to the limit you set
• Honest about limits: a nonexistent repo returns a clear _error record (even though the platform itself replies with a slightly confusing HTTP 401 instead of 404 — this actor normalizes that), and user vs. organization lookups are resolved automatically

How it works

Pick a mode:

1. models / datasets / spaces — search/browse with filters for query text, author, tags, task (models), library (models), sort order.
2. model-detail / dataset-detail / space-detail — fetch full detail (every file, card metadata, config) for one or more specific repo IDs.
3. creator — look up a user or organization profile (auto-detects which type).

Input

Search models:

{
  "mode": "models",
  "search": "llama",
  "pipelineTag": "text-generation",
  "sort": "downloads",
  "maxItems": 50
}

Model detail (by ID):

{
  "mode": "model-detail",
  "modelIds": ["google-bert/bert-base-uncased", "meta-llama/Llama-3.1-8B-Instruct"]
}

Datasets / Spaces:

{ "mode": "datasets", "search": "squad", "maxItems": 20 }

{ "mode": "spaces", "author": "huggingface", "sort": "likes", "maxItems": 20 }

Creator profile:

{ "mode": "creator", "username": "openai" }

Field	Type	Description
mode	string	"models" (default), "model-detail", "datasets", "dataset-detail", "spaces", "space-detail", or "creator".
search	string	models/datasets/spaces mode — free-text search term.
author	string	models/datasets/spaces mode — exact username/organization filter.
filter	array	models/datasets/spaces mode — tag filters, e.g. ["license:mit", "pytorch"].
pipelineTag	string	models mode — task filter, e.g. "text-generation".
library	string	models mode — e.g. "transformers", "diffusers".
sort	string	downloads, likes, createdAt, lastModified, or trendingScore.
direction	string	"-1" (descending, default) or "1" (ascending — only supported for createdAt, see Known limits).
modelId / modelIds	string / array	model-detail mode — repo ID(s) in namespace/name form.
datasetId / datasetIds	string / array	dataset-detail mode.
spaceId / spaceIds	string / array	space-detail mode.
username / usernames	string / array	creator mode.
accountType	string	creator mode — "auto" (default), "user", or "organization".
maxItems	integer	Max results for list modes. Default 100, max 10,000.
maxConcurrency	integer	Detail/creator modes — how many IDs to fetch in parallel. Default 4.
proxyConfiguration	object	Apify Proxy settings. Residential is on by default (not required — Hugging Face's own official public API — kept for production robustness).

Output

Models / model-detail mode — one record per model (truncated for readability):

{
  "_input": "google-bert/bert-base-uncased",
  "_source": "S1-api",
  "_scrapedAt": "2026-08-25T19:50:00Z",
  "id": "google-bert/bert-base-uncased",
  "author": "google-bert",
  "pipeline_tag": "fill-mask",
  "library_name": "transformers",
  "downloads": 93831528,
  "likes": 2740,
  "tags": ["transformers", "pytorch", "bert", "fill-mask", "license:apache-2.0"],
  "cardData": { "language": "en", "license": "apache-2.0" },
  "siblings": [{ "rfilename": "config.json" }, { "rfilename": "pytorch_model.bin" }]
}

Datasets / Spaces mode — analogous shape, upstream fields passed through as-is.

Creator mode:

{
  "_input": "openai",
  "_source": "S1-api-organization",
  "_scrapedAt": "2026-08-25T19:50:00Z",
  "fullname": "OpenAI",
  "name": "openai",
  "isVerified": true,
  "numModels": 39,
  "numDatasets": 16,
  "numSpaces": 6,
  "numFollowers": 40018
}

A record with _error instead of the fields above means that item could not be processed (for example, "_error": "model_not_found" for a repo ID that doesn't exist).

Known limits

• Ascending sort only works for createdAt. Requesting direction: "1" with any other sort value returns a structured _error: "invalid_request" record — this is a genuine Hugging Face platform restriction (confirmed live across all 5 sort fields), not a bug in this actor.
• List-mode results don't include the full file list. Only model-detail/dataset-detail/space-detail mode returns the complete siblings file manifest for a repo.
• Gated/private repos return their public metadata only (this actor never logs in, by design) — download-restricted file contents are out of scope.