docker-entrypoint.sh

#!/bin/sh
set -e

# Two deployment modes share this entrypoint (see DEPLOY.md):
#   - Local Docker  : OLLAMA_ENABLED=true  -> wait for Ollama + pull the model
#   - Hosted free web: OLLAMA_ENABLED=false -> start the server directly
# Hosted mode has NO Ollama service, so the wait/pull below must be skipped or
# the container would hang forever waiting for a server that never appears.

PORT="${PORT:-8000}"

case "$(echo "${OLLAMA_ENABLED:-true}" | tr '[:upper:]' '[:lower:]')" in
  0|false|no)
    echo "OLLAMA_ENABLED=false — hosted mode, skipping Ollama. Starting server…"
    exec uvicorn web.app:app --host 0.0.0.0 --port "${PORT}"
    ;;
esac

# ── Local Docker mode: bring up Ollama, then the server ──────────────────────
OLLAMA_URL="${OLLAMA_URL:-http://ollama:11434}"
MODEL="${DEFAULT_OLLAMA_MODEL:-llama3.2}"

echo "Waiting for Ollama at ${OLLAMA_URL}…"
until curl -sf "${OLLAMA_URL}/api/tags" > /dev/null 2>&1; do
  sleep 2
done
echo "Ollama ready."

# Pull the default model on first run if not already cached
if ! curl -sf "${OLLAMA_URL}/api/tags" | grep -q "\"${MODEL}\""; then
  echo "Pulling ${MODEL} (one-time download, ~2 GB)…"
  curl -sf "${OLLAMA_URL}/api/pull" \
    -H "Content-Type: application/json" \
    -d "{\"name\":\"${MODEL}\"}" \
    --no-buffer -o /dev/null
  echo "${MODEL} ready."
fi

exec uvicorn web.app:app --host 0.0.0.0 --port "${PORT}"