# Plexi API

FastAPI RAG backend deployed on HuggingFace Spaces.

## What It Does

- Loads the pre-built LlamaIndex vector store from the `plexi-materials` GitHub repo at startup
- Exposes three endpoints consumed by the Cloudflare Worker middleman

## Endpoints

| Method | Path | Purpose |
|---|---|---|
| `GET` | `/health` | Liveness probe — used by keep-alive GitHub Actions |
| `GET` | `/manifest` | Proxies + caches `manifest.json` from the materials repo |
| `POST` | `/retrieve` | Embeds query, searches index, returns scoped top-k chunks |

## Local Development

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

Visit `http://localhost:7860/docs` for the interactive API docs.

## Environment Variables

| Variable | Default | Purpose |
|---|---|---|
| `MATERIALS_REPO` | `KunalGupta25/plexi-materials` | GitHub repo with study materials |
| `MANIFEST_BRANCH` | `main` | Branch that holds `manifest.json` and `index/` |
| `ALLOWED_ORIGINS` | (Cloudflare Pages URL) | CORS allowed origins |

## Deploy to HuggingFace Spaces

1. Create a new Space with **Docker** SDK
2. Push this folder as the Space repo
3. Set environment variables in the Space settings
4. HF will build and run the Dockerfile automatically