scripts/check_llm_setup.sh

#!/usr/bin/env bash
# Verify hybrid LLM setup: Groq key, Ollama phi3.5 (routing), Qwen embed + fallback.
set -euo pipefail

ROOT="$(cd "$(dirname "$0")/.." && pwd)"
if [[ -f "$ROOT/.env" ]]; then
  set -a
  # shellcheck disable=SC1091
  source "$ROOT/.env"
  set +a
fi

BASE_URL="${GW_OLLAMA_BASE_URL:-${OLLAMA_BASE_URL:-http://127.0.0.1:11434}}"
ROUTING_MODEL="${GW_OLLAMA_ROUTING_MODEL:-phi3.5:latest}"
FALLBACK_MODEL="${GW_OLLAMA_SYNTHESIS_FALLBACK_MODEL:-qwen3.5:9b}"
EMBED_MODEL="${GW_OLLAMA_EMBEDDING_MODEL:-qwen3-embedding:8b}"
GROQ_MODEL="${GW_GROQ_MODEL:-${GROQ_MODEL:-llama-3.3-70b-versatile}}"
GROQ_KEY="${GW_GROQ_API_KEY:-${GROQ_API_KEY:-}}"

echo "Hybrid LLM check"
echo "  Ollama:     ${BASE_URL}"
echo "  Routing:    ${ROUTING_MODEL}  (Self-RAG + CRAG)"
echo "  Synthesis:  groq:${GROQ_MODEL} → fallback ollama:${FALLBACK_MODEL}"
echo "  Embedding:  ${EMBED_MODEL}"

TAGS_JSON="$(curl -sf "${BASE_URL%/}/api/tags" || true)"
if [[ -z "${TAGS_JSON}" ]]; then
  echo "ERROR: Ollama not reachable. Run: ollama serve"
  exit 1
fi

export TAGS_JSON ROUTING_MODEL FALLBACK_MODEL EMBED_MODEL
python3 <<'PY'
import json, os, sys
names = [m["name"] for m in json.loads(os.environ["TAGS_JSON"]).get("models", [])]
for label, model in [
    ("routing", os.environ["ROUTING_MODEL"]),
    ("synthesis fallback", os.environ["FALLBACK_MODEL"]),
    ("embedding", os.environ["EMBED_MODEL"]),
]:
    if model not in names:
        print(f"ERROR: Missing Ollama model for {label}: {model}")
        print(f"       ollama pull {model}")
        sys.exit(1)
    print(f"OK: {label} → {model}")
PY

if [[ -z "${GROQ_KEY}" ]]; then
  echo "WARNING: GROQ_API_KEY not set — synthesis will use Ollama fallback only."
else
  echo "OK: GROQ_API_KEY is set (groq:${GROQ_MODEL})"
fi

echo "Done."