Dockerfile

# ── Chatty: Dockerfile for Hugging Face Spaces (Docker SDK) ──
# Free tier: 2 vCPU, 16 GB RAM, 50 GB ephemeral disk, port 7860.
# Secrets are injected via HF Space settings, not .env files.

FROM python:3.14-slim AS base

# ── System dependencies ──
RUN apt-get update && apt-get install -y --no-install-recommends \
        ffmpeg \
        espeak-ng \
    && rm -rf /var/lib/apt/lists/*

# ── uv for fast dependency resolution ──
COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv

WORKDIR /app

# ── Python dependencies (cached layer — only invalidated when lock changes) ──
COPY pyproject.toml uv.lock ./
RUN uv sync --no-dev --frozen

# ── Pre-download the wav2vec2 phoneme model at build time ──
# This avoids a 5-10s cold-start delay on the first voice request.
# The model is ~1.2 GB and cached in the container image layer.
RUN uv run python -c "\
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor; \
Wav2Vec2Processor.from_pretrained('facebook/wav2vec2-lv-60-espeak-cv-ft'); \
Wav2Vec2ForCTC.from_pretrained('facebook/wav2vec2-lv-60-espeak-cv-ft'); \
print('wav2vec2 model cached')"

# ── Application code ──
COPY app/ app/
COPY alembic.ini ./

# ── HF Spaces requires port 7860 ──
ENV PORT=7860
EXPOSE 7860

# ── Writable paths: HF Spaces only allows /tmp for writes ──
# DATABASE_URL is intentionally NOT hardcoded here: production MUST set it to an
# external Postgres (Neon/Supabase) via HF Space Settings → Variables/Secrets, so
# data survives container restarts. Without it the app falls back to the SQLite
# default in app/core/config.py — fine for local dev, but on HF /tmp is wiped on
# every restart, so a prod Space left on SQLite loses all user data.
ENV AUDIO_DIR="/tmp/media/audio"
ENV AUDIO_STORAGE_BACKEND="local"

# ── Startup: run migrations then launch uvicorn ──
CMD ["sh", "-c", "python -c 'import os; print(\"Env keys:\", sorted(os.environ.keys()))' && uv run alembic upgrade head && uv run uvicorn app.main:app --host 0.0.0.0 --port 7860"]