Autosearch huggingface_hub

Discover open machine learning models on Hugging Face Hub via the public model search API.

install

source · Clone the upstream repo

git clone https://github.com/0xmariowu/Autosearch

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/0xmariowu/Autosearch "$T" && mkdir -p ~/.claude/skills && cp -r "$T/autosearch/skills/channels/huggingface_hub" ~/.claude/skills/0xmariowu-autosearch-huggingface-hub && rm -rf "$T"

manifest: autosearch/skills/channels/huggingface_hub/SKILL.md

source content

Overview

Hugging Face Hub provides broad coverage of open machine learning models through a free public search endpoint. It is useful when the query is trying to identify specific models, compare popular model families, or find research-adjacent model artifacts by task, library, and ecosystem tags.

When to Choose It

Choose it for model lookup queries like LLM families, embedding models, rerankers, vision models, and diffusion checkpoints.
Choose it when download counts, likes, pipeline type, and Hub tags are useful ranking signals even if the list endpoint does not expose long descriptions.
Choose it when the search should stay free and no-auth while still targeting the Hugging Face ecosystem directly.

How To Search

```
api_search
```
- Calls
```
https://huggingface.co/api/models
```
with
```
search=<query>
```
and
```
limit=10
```
, then maps public model hits into normalized evidence.
```
api_search
```
- Uses the model
```
id
```
as both canonical title and URL suffix, producing links like
```
https://huggingface.co/<id>
```
.
```
api_search
```
- Synthesizes snippet text from
```
pipeline_tag
```
,
```
library_name
```
, downloads, likes, and the first five tags because the list endpoint does not provide free-text summaries.

Known Quirks

Private or gated models are filtered client-side by skipping items where
```
private=True
```
.
The list endpoint returns no prose description, so snippets are synthesized from tags, task type, library, and popularity metadata.
Download and like counts can exceed 1M for popular models, so both are formatted with thousand separators for readability.

Quality Bar

Evidence items have non-empty title and url.
No crash on empty or malformed API response.
Source channel field matches the channel name.