README.md

---
title: G.U.I.D.E.
emoji: 📋
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false
---

# G.U.I.D.E.
G.U.I.D.E. - Grievance Utility for Information extraction, Drafting and Enrichment

## Deploying to Hugging Face Spaces

### 1. Upload model weights to HF Model Hub (one-time)

```bash
huggingface-cli login
python scripts/upload_models_to_hub.py --repo-id <your-username>/guide-models
```

This uploads only inference-needed files (`config.json`, `model.safetensors`, `tokenizer*.json`, `model.pt`). Training checkpoints are excluded automatically.

### 2. Create a Dockerfile Space on HF

Create a new Space at huggingface.co/spaces with **Dockerfile** as the Space SDK.

### 3. Push the code

```bash
git remote add hf https://huggingface.co/spaces/<your-username>/guide
git push hf main
```

### 4. Set the build argument

In the Space **Settings → Build**, set:
```
HF_MODEL_REPO=<your-username>/guide-models
```

### 5. Set the API key secret

In the Space **Settings → Secrets**, add:
```
ANTHROPIC_API_KEY=sk-ant-...
```

The key is injected at runtime — it is never baked into the Docker image.

### Architecture note

The Space runs two processes inside one container:
- **FastAPI** on `:8000` — internal only, not publicly exposed
- **Gradio** on `:7860` — the public-facing UI served by HF Spaces

The Gradio frontend calls `http://localhost:8000` internally; no ports other than 7860 need to be open.