# Publishing UnCorrupt to a HuggingFace Space

The `huggingface_space/` directory is the entire Space root. Pushing it to a HuggingFace Spaces git remote will trigger a build and live URL.

## One-time setup

```sh
pip install --upgrade huggingface_hub
huggingface-cli login   # paste a write-scoped token from https://huggingface.co/settings/tokens

# Create the Space (Docker SDK):
huggingface-cli repo create uncorrupt --type space --space_sdk docker
```

## Push the Space contents

```sh
cd huggingface_space
git init
git lfs install
git remote add origin https://huggingface.co/spaces/Sherpa/uncorrupt
git add README.md Dockerfile HOWTO.md
git commit -m "UnCorrupt v1.0.0 Space : Gradio detection + schema sidecar"
git push origin main
```

The Space build runs the Dockerfile (~10 min first time as deps install).

## Updating

Each push to the Space remote triggers a fresh build. Pinning a specific UnCorrupt release:

```sh
docker build --build-arg UNCORRUPT_REF=v1.0.0 -t test:local .
```

In the Space's Dockerfile, change `ARG UNCORRUPT_REF=main` to a tag once published.

## Free-tier defaults (2026)

- CPU-basic hardware (2 vCPU, 16 GB RAM) : sufficient for spreadsheet scanning
- 1 GB persistent storage (we don't need it; files are processed in-memory)
- HTTPS endpoint at `https://llmsherpa-uncorrupt.hf.space`
- Auto-sleep after 48h inactivity; first request after sleep cold-starts (~30s)

## Privacy notice for users

The Space card README declares: files are processed in the container and discarded. To be fully honest, that's accurate but the container DOES exist on HuggingFace infra. For users with truly sensitive data (clinical genomics, pre-publication), point them at the Pyodide browser-only variant where the file never leaves their browser.