Dockerfile.jetson

# Jetson Orin Dockerfile (JetPack 6 / L4T)
# For 16GB VRAM: 14B + 7B config
# For 8GB VRAM: Use zeta-lite profile

FROM nvcr.io/nvidia/l4t-cuda:12.2.2-runtime

# Install runtime dependencies
RUN apt-get update && apt-get install -y \
    curl \
    ca-certificates \
    python3 \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# Copy the pre-built aarch64 binary
# Build with: cmake -DGGML_CUDA=ON && make zeta-server
COPY dist/zeta-server-aarch64 /app/zeta-server
RUN chmod +x /app/zeta-server

# Create directories for models, storage, and dreams
RUN mkdir -p /models /storage /storage/dreams/pending /storage/dreams/archive /storage/blocks

# Environment variables (14B + 7B production config)
ENV ZETA_HOST="0.0.0.0"
ENV ZETA_PORT="8080"
ENV MODEL_MAIN="/models/qwen2.5-14b-instruct-q4_k_m.gguf"
ENV MODEL_CODER="/models/qwen2.5-coder-7b-instruct-q4_k_m.gguf"
ENV MODEL_EMBED="/models/nomic-embed-text-v1.5.f16.gguf"
ENV ZETA_STORAGE="/storage"
ENV ZETA_DREAM_INTERVAL="300"
ENV ZETA_DREAM_DIR="/storage/dreams"
ENV ZETA_GRAPH_PATH="/storage/blocks/graph.bin"

# GPU layers for Jetson Orin (16GB unified memory)
ENV GPU_LAYERS_MAIN="49"
ENV GPU_LAYERS_CODER="35"

EXPOSE 8080

COPY docker-entrypoint.sh /app/
COPY scripts/index_codebase.py /app/scripts/
RUN chmod +x /app/docker-entrypoint.sh

HEALTHCHECK --interval=30s --timeout=10s --start-period=120s --retries=3 \
    CMD curl -f http://localhost:8080/health || exit 1

ENTRYPOINT ["/app/docker-entrypoint.sh"]