Read what the API will not tell you

Parts of the hub have no API route at all, and other parts tell a browser more than they tell /api. The rendered pages carry their data as JSON in data-target and data-props attribute pairs, which is how the site hydrates itself in the browser. hf reads those pairs rather than scraping text, so what comes back is the site's own structured data, not a guess about its HTML.

The whole page as JSON

hf page google-bert/bert-base-uncased
hf page datasets/rajpurkar/squad -o json | jq '.[0].components | keys'
hf page google -o json | jq '.[0].components.OrgProfile'

hf page is 1:1 with what the browser gets. The record carries the page's title, its canonical URL, its meta tags, and a components map holding every declared component with its props underneath. If you are trying to work out whether a field exists anywhere, start here.

Merging it into a record

--deep on any read command fetches the page as well as the API and merges the fields only the page carries:

hf model google-bert/bert-base-uncased --deep
hf dataset rajpurkar/squad --deep
hf org google --deep

What you gain

With --deep	Why the API cannot
tagObjs, decoded tags with human labels and sub-types	The API returns raw tag strings
open and closed discussion counts	No API route
linked spaces with their live running state	The API's spaces expand is ids only
an org's models, datasets, spaces, collections, members, and papers	Seven paginated calls otherwise
paper AI summaries, keywords, and upvoters	No API route
the dataset viewer's rendered rows and facet counts	Different shape
a space's runtime state inline	Saves the separate /runtime call

An org page is the clearest win: one 254 KB payload replaces seven paginated request sequences.

What it costs

A page is roughly ten times the bytes of the equivalent API call, and it is a second request. hf reaches for it only when asked, and never when a dedicated endpoint returns the same field. On a crawl of thousands of repos, leaving --deep off is usually the right call.

Commands that are page-only

Some of the surface has no API behind it at all, so these read the page whichever way you ask:

• hf upvoters for papers and collections
• hf post for a single community post
• hf blog for a blog entry's body
• parts of hf paper, including the AI summary

They behave like every other command. The page plane is an implementation detail, and the only place it shows through is the sources field on the record, which lists exactly what was fetched to build it.