Hugging Face AI Org Lead Extractor

Turn Hugging Face model authors into ready-to-pitch AI-company sales leads.

For each AI org, this Actor discovers models via Hugging Face's public keyless Hub API, confirms the namespace is an organization (not an individual — a ToS-safe gate), discovers the org's own company website (via its HF org page + the top model's model-card README), and enriches that site with the same MX/DNS-verified email + phone + social + tech-stack + A–D lead-score layer used across the API Factory lead-extractors. Each lead also carries org firmographics (followers, model count, verified badge) and a trending/growing-AI-org signal (HF traction + publishing recency + contactability → A–D grade).

No HF author personal emails are mined (that is the ToS-gray personal-data harvest we reject). The contact always comes from the AI org's own company website.

No API keys. No residential proxy. No compose cost. Works straight out of the box — Hugging Face's Hub API is public and keyless, and we fetch each org's own public company site directly. Margin-positive from day one.

---

Use cases

• B2B vendor prospecting — find AI orgs in a space you sell into (GPU/cloud, eval, inference, observability, data, dev-tool, MLOps vendors) with a verified contact on their own company site.
• Investor / partnership outreach — surface trending/growing AI orgs (high HF followers + recent model publishes + reachable contact) before they're widely known.
• Monitoring / "newly published since last run" — schedule with sinceDays to get only AI orgs that published a model in the trailing N days.

How it works

1. Discover models via huggingface.co/api/models?search=…&sort=likes (or pipeline_tag= for known HF tasks like text-generation), newest-first when a recency filter is set. Models are rolled up to one lead per org namespace (the slug before /).
2. Org gate (ToS-safe) — /api/organizations/{ns}/overview returns 200 for an organization and 404 for an individual user. Only org namespaces are kept; individuals are skipped (never mined). The overview also yields firmographics (fullname, isVerified, numModels, numFollowers, numPapers).
3. Company-website hop — the org's declared external link on its HF org page is the authoritative company website (e.g. mistral.ai, bfl.ai, deepseek.com). If the org page has no external site, the model-card README is scanned and a link is trusted only if its apex domain matches the org namespace — so a third-party paper/blog/license site is never misattributed.
4. Enrich the company website with MX/DNS-verified emails, phones, socials, tech stack, and an A–D lead score (shared enrich.py layer).
5. Bill only per AI-org lead that returns a real contact signal — empty / no-contact orgs are returned honestly and never billed.

Input

Field	Type	Default	Description
searchQueries	string[]	["text-generation"]	Keywords or HF pipeline tags (e.g. text-generation, text-to-image, AI agent). Pipeline tags use the exact task filter (surfaces real orgs by traction); freeform keywords use full-text search.
orgSlugs	string[]	[]	Explicit HF org namespace slugs to enrich directly (e.g. mistralai), bypassing keyword search.
maxPerQuery	int	20	Cap on models per query (1–100); these roll up to one lead per org.
sinceDays	int	0	Keep only models published in the trailing N days. 0 = all time. Monitoring mode (sorts newest-first).
minLikes	int	0	Minimum HF likes on the org's representative model (community traction).
verifyEmails	bool	true	MX/DNS-validate the best contact email's domain.
enrichWebsite	bool	true	Discover the org's company site + extract emails/phones/socials/tech. Disable for metadata-only runs (no contact data, nothing billed).
concurrency	int	5	Parallel orgs enriched (1–10).
proxyConfiguration	object	{}	Optional — only if you hit rate limits.

Provide searchQueries and/or orgSlugs. A typical run takes ~20–40s for ~20 orgs at the default concurrency.

Output

One dataset item per AI org (organizations only; individuals are not emitted as billed leads):

{
  "namespace": "deepseek-ai",
  "modelSlug": "DeepSeek-R1",
  "modelUrl": "https://huggingface.co/deepseek-ai/DeepSeek-R1",
  "pipelineTag": "text-generation",
  "downloads": 8739585,
  "likes": 13461,
  "modelCount": 5,
  "newestModelDate": "2026-04-22T06:04:45+00:00",
  "orgUrl": "https://huggingface.co/deepseek-ai",
  "isOrg": true,
  "orgFullname": "DeepSeek",
  "isVerified": true,
  "numFollowers": 137393,
  "numModels": 101,
  "numPapers": 27,
  "companyUrl": "https://www.deepseek.com/",
  "emails": ["service@deepseek.com", "security@deepseek.com"],
  "emailVerified": true,
  "phones": [],
  "socials": { "twitter": "https://twitter.com/deepseek_ai" },
  "techStack": ["React"],
  "orgReadiness": "A",
  "orgLabel": "Prime trending AI-org target — strong HF traction, recently publishing, reachable contact.",
  "leadScore": "A",
  "hasContact": true
}

The sample above is a real lead from a live run (

searchQueries: ["text-generation"]

, 2026-07-13) — not fabricated. Fields that have no public value return [] / null honestly (e.g. phones: [] here) rather than placeholder data.

Field reference

Field	Meaning
namespace	The HF org slug (mistralai, deepseek-ai). One lead per org.
isOrg	true if the overview endpoint confirmed an organization; false = individual (skipped, not billed).
orgFullname / isVerified / numFollowers / numModels / numPapers	Org firmographics from the HF overview.
modelSlug / modelUrl / pipelineTag / downloads / likes	The org's representative (highest-likes) matching model.
companyUrl	The org's own company website (discovered, trusted). null if none found.
emails / emailVerified	MX/DNS-verified business contact emails from the org's own site.
socials / techStack	Social profiles + detected tech stack from the org's own site.
orgReadiness / orgLabel	A–D trending-org grade (followers + model traction + publishing recency + contactability).
leadScore	A–D signal-richness grade (email+verified+phone+social+tech).
hasContact	true if any contact signal was found — the billable event.

Pricing (pay-per-event)

You are charged only for successful results — never for empty/no-contact runs (the #1 market complaint about flat-rate metadata-dump scrapers).

Event	Price	When charged
Actor Start	$0.0001	Once per run start.
Verified AI-Org Lead (primary)	$0.008	Once per AI org that returns ≥1 real contact signal.
Verified Org Email	$0.003	When an org yields ≥1 MX/DNS-validated email.
Trending-Org Signal	$0.001	When the A–D org-readiness grade is computed.

A typical billed lead (verified email + readiness grade) costs ~$0.012. With zero upstream compose cost (keyless HF API + direct fetch of the org's own site), each result is margin-positive from the first run.

Performance & limits

• ~20–40s for 20 orgs at default concurrency on Apify.
• The HF Hub API is keyless and generous; a proxy is rarely needed. If you hit rate limits on very large sweeps, set proxyConfiguration.
• Some orgs (research groups, e.g. bigscience) have no public company website on their HF page — these return companyUrl: null and hasContact: false and are not billed (honest no-contact, not a fabricated lead).

Troubleshooting

• An org I expected is missing — it may be an individual user namespace (isOrg: false, skipped per the ToS-safe gate) or it had no matching model in the top maxPerQuery results. Raise maxPerQuery or pass it directly via orgSlugs.
• companyUrl is null — the org's HF page and model-card README exposed no external company site, or only third-party/paper links that failed the name-match trust check. The org degrades to metadata-only and is not billed.
• emails is [] but hasContact is true — the org's site had socials (or phones) but no public email; that still counts as a reachable contact.
• Want faster, metadata-only runs — set enrichWebsite: false (no contact data, nothing billed; returns org firmographics + model metadata only).

Data sources & compliance

• Hugging Face public Hub API (/api/models, /api/organizations/{ns}/overview) — keyless, no anti-bot.
• Each AI org's own public company website — fetched directly (no platform pages: GitHub/Twitter/etc. are skipped via the shared platform-host gate).
• No HF author/individual personal data is mined; contact is always sourced from the organization's own website.