Dockerfile

FROM node:20-slim

# System deps: python3 + build tools for llama-cpp-python + curl for model download
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 python3-pip make g++ curl unzip \
    && rm -rf /var/lib/apt/lists/*

# Install llama-cpp-python server — try prebuilt first, compile if needed
# CPU-only build (GGML_NATIVE=0 disables AVX/AVX2 for max compatibility)
RUN CMAKE_ARGS="-DGGML_NATIVE=OFF" FORCE_CMAKE=1 \
    pip3 install --break-system-packages --no-cache-dir \
    "llama-cpp-python[server]" \
    --prefer-binary \
    --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu \
    && echo "✅ llama-cpp-python installed!" \
    || echo "⚠️ llama-cpp-python install failed — local model unavailable"

WORKDIR /app
RUN mkdir -p /data /app/models && chmod 777 /data

# npm install (Node server deps)
COPY package.json ./
RUN npm install

# Pre-download TinyLlama 1.1B Q4_K_M GGUF (~637MB)
# Best effort — won't fail build if download fails
RUN curl -L --max-time 600 --retry 3 \
    "https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf" \
    -o /app/models/local.gguf \
    && echo "✅ TinyLlama $(du -sh /app/models/local.gguf | cut -f1)" \
    || echo "⚠️ Model download failed — will download at first startup"

# Copy pre-built Vite dist + server
COPY dist/ ./dist/
COPY server.js ./

RUN chown -R 1000:1000 /app && chmod 777 /data
USER 1000

EXPOSE 7860
ENV PORT=7860 NODE_ENV=production

CMD ["node", "server.js"]