Hugging Face
The live Hub connection behind model discovery, creator profiles, documentation, and sign-in.
6 sections / hugpad handbook
On this page
Built around the Hugging Face Hub
The Hugging Face Hub is hugpad’s source for public models and their publishers. The Hub hosts versioned model repositories, datasets, and Spaces. hugpad focuses on public model discovery and creator profiles.
Live integration
What hugpad reads
| Hub information | Where it is used |
|---|---|
| Model IDs and task metadata | Directory search, categories, and links to source repositories. |
| Model counts and dates | Reported downloads, likes, trending scores, creation dates, and updates when supplied. |
| README and repository metadata | Model pages, documentation, license labels, and tags. |
| Recent commits | Model change history. |
| Profiles and publishers | Creator names, avatars, biographies, account types, and available counts. |
| Signed-in identity and organization roles | Server-side verification of who can launch for a repository namespace. |
Sign-in and permissions
Public browsing does not need sign-in. To launch, hugpad redirects you to Hugging Face’s OAuth sign-in and uses PKCE to protect the sign-in exchange. The requested scopes are openid and profile.
hugpad does not ask you to paste a Hugging Face access token into the launch form or upload model weights. The browser stores the sign-in session; expired sessions require a fresh sign-in. Launch operations recheck the identity against Hugging Face on the server.
See ownership verification for organization-role requirements and security for session handling.
Freshness and missing data
Short-lived query caches reduce repeated Hub requests. Model queries are typically fresh for five minutes, so a change upstream may not appear immediately. Directory requests follow real next-page links rather than constructing pretend pages.
Not every repository publishes all metadata. A missing license, size, count, or description is not filled with an invented value. A Hub outage or rate limit may temporarily interrupt browsing without changing what the underlying model is.
What this integration does not do
- It does not host or execute every model listed on the Hub.
- It does not bypass gated or private model access.
- It does not grant model rights through token ownership.
- It does not turn a Hub verification badge into launch authorization.
- It does not promise that every model’s reported evaluation or statistics have been independently checked.