---
title: AirGPT Hub
colorFrom: green
colorTo: gray
sdk: docker
app_port: 7860
pinned: false
license: mit
---

# AirGPT Hub

Vertical model and dataset discovery for embodied AI. The Space reads the public
[`airgpt/embodied-model-index`](https://huggingface.co/datasets/airgpt/embodied-model-index)
catalog and exposes a responsive browser plus a small API.

The interface shows verified runtime revisions, VRAM guidance, model licenses, real
validation previews, and copyable AirGPT Runtime commands. Asset provenance is recorded
in [`ASSET_NOTICES.md`](ASSET_NOTICES.md).

The Explore catalog separates runtime-verified models from community models whose AirGPT
adapters are still pending. It also includes a curated robotics dataset browser. Entries
are enriched with cached public Hugging Face metadata, including revision drift,
downloads, likes, gating, licenses, and storage. The suggestion form validates public
model repositories and generates an unverified candidate manifest. It does not store
submissions or hold an organization token.

## Local development

```bash
python -m pip install -r requirements-dev.txt
uvicorn app.main:app --reload --port 7860
pytest
```

Set `AIRGPT_CATALOG_URL` to use another catalog. If Hugging Face is unavailable, the app
falls back to `data/index.json` bundled in the image.

The organization Space uses the Docker SDK with FastAPI, a live five-minute catalog
and metadata cache. The synchronized `data/index.json` snapshot remains available
when Hugging Face is temporarily unreachable.

`index.json` remains the strict AirGPT Runtime contract. Web-only community models live in
`community-models.json`, while curated robotics datasets live in `datasets.json`. This
keeps `airgpt hub search` and `airgpt pull` compatible with verified runtime adapters.

Maintainers can review a candidate locally. The command prints a merged preview unless
`--publish` is explicitly supplied:

```bash
python scripts/review_candidate.py candidate.json
HF_TOKEN=... python scripts/review_candidate.py candidate.json --publish
```

Publish reviewed community and dataset catalogs explicitly:

```bash
python scripts/publish_catalogs.py
HF_TOKEN=... python scripts/publish_catalogs.py --publish
```

## Publish

```bash
export HF_TOKEN=...
HTTPS_PROXY=http://127.0.0.1:7890 python scripts/publish_space.py
```