Quick start

Read your first record, then walk the graph.

Once hf is on your PATH, read a model:

hf model google-bert/bert-base-uncased

The argument is a ref, and every command that takes one accepts every form of it. A bare name, an owner-qualified id, a URL you copied out of a browser, and a canonical URI are all the same command:

hf model bert-base-uncased
hf model google-bert/bert-base-uncased
hf model https://huggingface.co/google-bert/bert-base-uncased
hf model hf://model/google-bert/bert-base-uncased

If you do not know what kind of thing a ref is, hf get works it out:

hf get rajpurkar/squad
hf get https://huggingface.co/collections/fdtn-ai/antares-6a5804c889e78b51c447c38a

Shape the output

The same flags work on every command:

hf model google-bert/bert-base-uncased --fields id,downloads,likes,pipeline_tag
hf model google-bert/bert-base-uncased -o json | jq .safetensors
hf models --author google -o url | head -5

-o takes table, json, jsonl, csv, tsv, url, or raw. Left to auto, it prints a table to a terminal and JSONL into a pipe, so the same command reads well by hand and parses cleanly downstream. See output formats for the full contract.

List and search

hf models --author google --task text-generation --sort downloads -n 20
hf datasets --search squad -n 10
hf search bert                     # every entity kind at once
hf trending --type model

Listing streams. -n stops early without fetching the next page, and leaving it off means the whole result set, which for hf models is a few million records and a very long afternoon.

Go deeper

The API does not return everything the page shows. --deep fetches the rendered page as well and merges the fields only it carries:

hf model google-bert/bert-base-uncased --deep -o json | jq .tagObjs
hf org google --deep -o json | jq '.models | length'

--card adds the README body, and hf card gives you just the parsed front matter.

Walk the graph

Every record knows what it points at:

hf graph google-bert/bert-base-uncased        # the node and its edges
hf edges google-bert/bert-base-uncased        # edges only
hf children google-bert/bert-base-uncased     # what was fine-tuned from it
hf crawl hf://org/google --depth 2 -n 500

See walk the graph for what the edges mean and where each one comes from.

Export it

hf rdf google-bert/bert-base-uncased --format ttl
hf export hf://org/google --depth 2 --format jsonl > google.jsonl
hf croissant rajpurkar/squad

See linked data for the vocabulary.

Serve it instead

The same operations are available over HTTP and to agents over MCP:

hf serve --addr :7777 &
curl -s 'localhost:7777/v1/model/google-bert/bert-base-uncased'   # NDJSON
hf mcp                                                            # MCP over stdio