Dockerfile

# Single-container Hugging Face Space image for 9XAIPal.
# The container starts PostgreSQL (with pgvector) + Redis in the background and
# then launches the plain FastAPI app (app.py) on port 7860 via supervisord.

# Pin to Debian bookworm: the PGDG apt repo below targets bookworm, and the
# default python:3.11-slim now resolves to Debian trixie (libicu76/libldap-2.6),
# whose libs can't satisfy the bookworm PostgreSQL packages.
FROM python:3.11-slim-bookworm

ENV PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    # Tell uvicorn to bind on 0.0.0.0 by default.
    HOST=0.0.0.0 \
    PORT=7860 \
    # Default paths inside the container.
    UI_DIR=/code/dist \
    # In-container Ollama (embeddings only): bind locally and store the baked
    # model where both build-time pull and runtime serve will look for it.
    OLLAMA_HOST=127.0.0.1:11434 \
    OLLAMA_MODELS=/root/.ollama/models \
    # The Space image does not bake MinerU (several GB). Enable the PyMuPDF
    # text-only fallback so uploads still process out of the box; install MinerU
    # and set this to 0 for higher-fidelity extraction.
    ALLOW_PYMUPDF_FALLBACK=1

WORKDIR /code

# Install Redis, PostgreSQL 16 + pgvector, and basic build tooling.
#   - Redis is required by the backend Celery queue.
#   - PostgreSQL (with the pgvector extension) is the app's primary database;
#     it runs inside this container and is created/initialized by start.sh.
#   - build-essential/libpq-dev are here in case a dependency wheel is
#     unavailable for the current platform.
# PostgreSQL + pgvector come from the official PGDG apt repository.
RUN apt-get update \
    && apt-get install -y --no-install-recommends \
        ca-certificates curl gnupg \
        zstd \
        redis-server \
        supervisor \
        build-essential \
        libpq-dev \
    && install -d /usr/share/postgresql-common/pgdg \
    && curl -fsSL https://www.postgresql.org/media/keys/ACCC4CF8.asc \
         -o /usr/share/postgresql-common/pgdg/apt.postgresql.org.asc \
    && echo "deb [signed-by=/usr/share/postgresql-common/pgdg/apt.postgresql.org.asc] https://apt.postgresql.org/pub/repos/apt bookworm-pgdg main" \
         > /etc/apt/sources.list.d/pgdg.list \
    && apt-get update \
    && apt-get install -y --no-install-recommends \
        postgresql-16 \
        postgresql-16-pgvector \
    && rm -rf /var/lib/apt/lists/*

# Install Ollama and bake the embedding model into the image. Chat runs on
# Ollama Cloud (a 31B model), but the cloud doesn't host embedding models, so we
# serve qwen3-embedding from an in-container Ollama at localhost:11434. Pulling
# the weights at build time bakes them into an image layer, so cold starts don't
# re-download ~640 MB and the Space works even without a persistent disk.
RUN curl -fsSL https://ollama.com/install.sh | sh \
    && (ollama serve & srv=$!; \
        for i in $(seq 1 30); do ollama list >/dev/null 2>&1 && break; sleep 1; done; \
        ollama pull qwen3-embedding; \
        kill "$srv" 2>/dev/null || true)

# Install Python dependencies first so layer caching works across code changes.
COPY requirements.txt .
RUN pip install --no-cache-dir --upgrade pip \
    && pip install --no-cache-dir -r requirements.txt

# Copy the rest of the project (backend code, tests, and the React dist build).
COPY . .

COPY supervisord.conf /code/supervisord.conf
RUN chmod +x /code/start.sh

EXPOSE 7860

# Docker will restart the container if /healthz starts returning unhealthy.
# start.sh brings up PostgreSQL (+pgvector), Redis and Ollama, then supervisord
# keeps the API + Celery worker alive.
HEALTHCHECK --interval=30s --timeout=15s --start-period=120s --retries=4 \
    CMD curl -fsS http://127.0.0.1:7860/healthz >/dev/null || exit 1

CMD ["bash", "/code/start.sh"]