scripts/download-models.sh

#!/bin/bash
# SolVox Model Downloader
# Downloads all required GGUF/ONNX models for QVAC SDK
# All models are stored locally — no cloud required after download.

set -e

MODELS_DIR="./models"
mkdir -p "$MODELS_DIR"

echo "╔══════════════════════════════════════════════════════════╗"
echo "║           SolVox — QVAC Model Downloader                ║"
echo "║    Downloading AI models for 100% local inference        ║"
echo "╚══════════════════════════════════════════════════════════╝"
echo ""

# 1. LLM — Llama 3.2 3B Instruct (Q4_K_M quantization, ~2GB)
echo "📦 [1/6] Downloading LLM: Llama 3.2 3B Instruct (Q4_K_M)..."
if [ ! -f "$MODELS_DIR/llama-3.2-3b-instruct-q4_k_m.gguf" ]; then
  curl -L -o "$MODELS_DIR/llama-3.2-3b-instruct-q4_k_m.gguf" \
    "https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF/resolve/main/Llama-3.2-3B-Instruct-Q4_K_M.gguf"
  echo "   ✓ LLM model downloaded (~2.0 GB)"
else
  echo "   ✓ LLM model already exists"
fi

# 2. Embeddings — Nomic Embed Text v1.5 (Q4_K_M, ~260MB)
echo "📦 [2/6] Downloading Embeddings: Nomic Embed Text v1.5..."
if [ ! -f "$MODELS_DIR/nomic-embed-text-v1.5.Q4_K_M.gguf" ]; then
  curl -L -o "$MODELS_DIR/nomic-embed-text-v1.5.Q4_K_M.gguf" \
    "https://huggingface.co/nomic-ai/nomic-embed-text-v1.5-GGUF/resolve/main/nomic-embed-text-v1.5.Q4_K_M.gguf"
  echo "   ✓ Embedding model downloaded (~260 MB)"
else
  echo "   ✓ Embedding model already exists"
fi

# 3. Speech-to-Text — Whisper Base English (GGML, ~150MB)
echo "📦 [3/6] Downloading STT: Whisper Base (English)..."
if [ ! -f "$MODELS_DIR/ggml-base.en.bin" ]; then
  curl -L -o "$MODELS_DIR/ggml-base.en.bin" \
    "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin"
  echo "   ✓ Whisper model downloaded (~150 MB)"
else
  echo "   ✓ Whisper model already exists"
fi

# 4. Text-to-Speech — Piper TTS Amy (en_US, medium quality, ONNX, ~75MB)
echo "📦 [4/6] Downloading TTS: Piper Amy (en_US, medium)..."
if [ ! -f "$MODELS_DIR/en_US-amy-medium.onnx" ]; then
  curl -L -o "$MODELS_DIR/en_US-amy-medium.onnx" \
    "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/amy/medium/en_US-amy-medium.onnx"
  curl -L -o "$MODELS_DIR/en_US-amy-medium.onnx.json" \
    "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/amy/medium/en_US-amy-medium.onnx.json"
  echo "   ✓ TTS model downloaded (~75 MB)"
else
  echo "   ✓ TTS model already exists"
fi

# 5. Translation — MarianMT English↔Spanish (OPUS, ~50MB)
echo "📦 [5/6] Downloading Translation: EN↔ES..."
if [ ! -f "$MODELS_DIR/translate-en-es.bin" ]; then
  echo "   ℹ Translation model requires manual download from LibreTranslate"
  echo "   Visit: https://github.com/LibreTranslate/LibreTranslate"
  echo "   Or use QVAC's bundled translation models (see docs.qvac.tether.io)"
  touch "$MODELS_DIR/translate-en-es.bin.placeholder"
  echo "   ⚠ Placeholder created — install actual model for translation"
else
  echo "   ✓ Translation model already exists"
fi

# 6. OCR — PaddleOCR v4 (ONNX, ~30MB)
echo "📦 [6/6] Downloading OCR: PaddleOCR v4..."
if [ ! -f "$MODELS_DIR/ppocr-v4.onnx" ]; then
  echo "   ℹ OCR model requires QVAC's bundled OCR models"
  echo "   See: https://docs.qvac.tether.io/addons/ocr-onnx/"
  touch "$MODELS_DIR/ppocr-v4.onnx.placeholder"
  echo "   ⚠ Placeholder created — install actual model for OCR"
else
  echo "   ✓ OCR model already exists"
fi

echo ""
echo "╔══════════════════════════════════════════════════════════╗"
echo "║                  Download Complete!                      ║"
echo "╚══════════════════════════════════════════════════════════╝"
echo ""
echo "Model directory: $MODELS_DIR"
echo ""
ls -lh "$MODELS_DIR"
echo ""
echo "Next steps:"
echo "  1. npm install"
echo "  2. npm run dev"
echo "  3. npm start"