feat: run qdrant and ollama inside thothii

scripts/run-stack.sh

@@ -1,8 +1,8 @@
 #!/usr/bin/env bash
 # run-stack.sh — avvia lo stack Compose locale di ThothII in primo piano.
 #
-# Il core include Pi; DWH, vector DB, embedding e LLM sono endpoint esterni configurati
-# in deploy/env/local.env. Non richiede un eseguibile Pi sull'host.
+# Il core include Pi; DWH e LLM restano endpoint esterni configurati in deploy/env/local.env.
+# Qdrant e Ollama embedding sono servizi Compose privati. Non richiede un eseguibile Pi sull'host.
 #
 # Preparazione: cp deploy/env/local.env.example deploy/env/local.env e compilare i valori.
 # Uso: ./scripts/run-stack.sh [argomenti aggiuntivi per docker compose up]
@@ -16,5 +16,10 @@ LOCAL_ENV_FILE="${THT_LOCAL_ENV_FILE:-$ROOT/deploy/env/local.env}"
   exit 1
 }
 
+compose_files=(-f "$ROOT/compose.yaml" -f "$ROOT/deploy/compose.local.yaml")
+if [[ "${THOTH_ENABLE_EMBEDDING_GPU:-0}" == "1" ]]; then
+  compose_files+=(-f "$ROOT/deploy/compose.embedding-gpu.yaml")
+fi
+
 exec docker compose --env-file "$LOCAL_ENV_FILE" \
-  -f "$ROOT/compose.yaml" -f "$ROOT/deploy/compose.local.yaml" up --build "$@"
+  "${compose_files[@]}" up --build "$@"