start.sh

#!/bin/bash
# Start the Docling + Gemini hybrid parser API.
#
# UVICORN_WORKERS: number of worker processes (default 2).
#   Each worker loads its own copy of the Docling model, so don't set this
#   higher than RAM allows. On T4 Small (15GB RAM), 2 is a safe default.
#   Set UVICORN_WORKERS=1 to revert to single-process mode.

exec uvicorn app:app \
  --host 0.0.0.0 \
  --port 7860 \
  --workers "${UVICORN_WORKERS:-2}"