# Quickstart — DictIA

## Prerequis communs

- Docker + Docker Compose V2
- Git
- 2GB+ RAM disponible

```bash
git clone https://gitea.innova-ai.ca/Innova-AI/dictia.git
cd dictia
git checkout dictia-branding
```

---

## Profil Cloud (VPS + GCP GPU)

Le GPU demarre automatiquement quand quelqu'un transcrit, et s'arrete apres 5 min d'inactivite.

```bash
# 1. Setup interactif
bash deployment/setup.sh --profile cloud

# 2. Setup ASR Proxy (GCP credentials requises)
bash deployment/asr-proxy/setup.sh

# 3. Optionnel: Tailscale Serve pour HTTPS
bash deployment/config/tailscale/setup-serve.sh
```

**Requis**: credentials GCP (service account ou OAuth) dans `deployment/asr-proxy/gcp-credentials.json`.

---

## Profil Local GPU

Transcription locale sur GPU NVIDIA. Le plus rapide.

```bash
# Prerequis: nvidia-container-toolkit
# https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html

# Setup
bash deployment/setup.sh --profile local-gpu
```

**Requis**: token HuggingFace pour la diarisation (pyannote).

---

## Profil Local CPU

Transcription sur CPU. Lent mais fonctionnel pour tester.

```bash
bash deployment/setup.sh --profile local-cpu
```

Prevoir ~10x le temps reel (1h audio = ~10h de traitement).

---

## Apres l'installation

```bash
# Verifier que tout fonctionne
bash deployment/tools/health-check.sh

# Ouvrir DictIA
open http://localhost:8899
```

Se connecter avec les identifiants admin configures pendant le setup.

## Commandes utiles

```bash
# Logs en temps reel
docker compose -f deployment/docker/docker-compose.<profil>.yml logs -f

# Redemarrer
docker compose -f deployment/docker/docker-compose.<profil>.yml restart

# Mise a jour
bash deployment/tools/update.sh

# Backup
bash deployment/tools/backup.sh
```