Deploying Rubriq
Two pieces: the engine (FastAPI + both models, Docker) on Hugging Face Spaces, and the static frontend on Vercel. The engine also serves the frontend itself, so the HF Space alone is a complete working app — the Vercel deploy just gives it a nicer URL.
1. Engine → Hugging Face Spaces
One-time setup, in any terminal:
# needs a free HF account; create a WRITE token at
# huggingface.co/settings/tokens
.venv\Scripts\pip install -U huggingface_hub
.venv\Scripts\hf auth login # paste the write token
# create the Space (Docker SDK) and push this repo to it
.venv\Scripts\python -c "from huggingface_hub import create_repo; create_repo('<HF_USERNAME>/rubriq', repo_type='space', space_sdk='docker')"
git remote add hf https://huggingface.co/spaces/<HF_USERNAME>/rubriq
git push hf main
The Space builds the Dockerfile (models bake into the image, ~10 min first
build) and serves at https://<HF_USERNAME>-rubriq.hf.space. Redeploys are
just git push hf main.
Free-tier note: Spaces sleep after ~48h idle; the first request after sleep takes ~1 min. The UI's status pill shows "waking the engine…" and retries by itself. Before the mentor demo, open the URL once in advance.
2. Frontend → Vercel
npx vercel login # once
cd web
npx vercel --prod --yes # deploys the static frontend
Then point the frontend at the engine — edit web/config.js:
window.RUBRIQ_API = "https://<HF_USERNAME>-rubriq.hf.space";
commit, and run npx vercel --prod --yes from web/ again.
Any deployed copy can also be re-pointed without a redeploy by opening
https://<frontend-url>/?api=https://<engine-url> once — the override
persists in that browser.
Local demo (no deployment at all)
.venv\Scripts\uvicorn rubriq.api.main:app --port 8000
# open http://127.0.0.1:8000 — UI and engine together, fully offline