Update .env

.env

@@ -171,30 +171,30 @@ GRAPH_ENTITY_EXTRACTION=false
 # ── Option 3 · HIGH · GPU · g6e.2xlarge · 8 cores · 64 GB RAM · 48 GB VRAM ───
 # Đang active. Mỗi model đều fit hoàn toàn trong 48 GB VRAM ở context 32K, nhưng
 # qwen3.6:35b + north-mini-code + KV cache không có dư địa an toàn để giữ đồng thời.
-# MAX_LOADED_MODELS=1 ưu tiên ổn định/chất lượng; đổi model chậm hơn nhưng tránh OOM.
-# OLLAMA_MAX_LOADED_MODELS=1
-# OLLAMA_NUM_PARALLEL=1
-# OLLAMA_CONTEXT_LENGTH=32768
-# OLLAMA_REQUEST_TIMEOUT=1200
-# RAG_TIMEOUT=900
-# CHAT_MODEL=qwen3.6:35b
-# EMBEDDING_MODEL=qwen3-embedding:8b
-# CODING_PLANNER_MODEL=granite4.1:8b
-# BA_MODEL=qwen3.6:35b
-# PM_MODEL=qwen3.6:35b
-# SA_MODEL=qwen3.6:35b
-# TA_MODEL=qwen3.6:35b
-# DESIGNER_MODEL=gemma4:31b
-# FE_MODEL=north-mini-code-1.0:q4_K_M
-# MOBILE_MODEL=north-mini-code-1.0:q4_K_M
-# BE_MODEL=north-mini-code-1.0:q4_K_M
-# DBA_MODEL=north-mini-code-1.0:q4_K_M
-# DA_MODEL=qwen3.6:35b
-# TL_MODEL=north-mini-code-1.0:q4_K_M
-# TECH_LEAD_MODEL=north-mini-code-1.0:q4_K_M
-# TESTER_MODEL=qwen3.6:35b
-# DEVSECOPS_MODEL=north-mini-code-1.0:q4_K_M
-# CLARIFIER_MODEL=qwen3.6:35b
+MAX_LOADED_MODELS=1 ưu tiên ổn định/chất lượng; đổi model chậm hơn nhưng tránh OOM.
+OLLAMA_MAX_LOADED_MODELS=1
+OLLAMA_NUM_PARALLEL=1
+OLLAMA_CONTEXT_LENGTH=32768
+OLLAMA_REQUEST_TIMEOUT=1200
+RAG_TIMEOUT=900
+CHAT_MODEL=qwen3.6:35b
+EMBEDDING_MODEL=qwen3-embedding:8b
+CODING_PLANNER_MODEL=granite4.1:8b
+BA_MODEL=qwen3.6:35b
+PM_MODEL=qwen3.6:35b
+SA_MODEL=qwen3.6:35b
+TA_MODEL=qwen3.6:35b
+DESIGNER_MODEL=gemma4:31b
+FE_MODEL=north-mini-code-1.0:q4_K_M
+MOBILE_MODEL=north-mini-code-1.0:q4_K_M
+BE_MODEL=north-mini-code-1.0:q4_K_M
+DBA_MODEL=north-mini-code-1.0:q4_K_M
+DA_MODEL=qwen3.6:35b
+TL_MODEL=north-mini-code-1.0:q4_K_M
+TECH_LEAD_MODEL=north-mini-code-1.0:q4_K_M
+TESTER_MODEL=qwen3.6:35b
+DEVSECOPS_MODEL=north-mini-code-1.0:q4_K_M
+CLARIFIER_MODEL=qwen3.6:35b
 
 # ── Option 4 · ULTRA · GPU · g6e.12xlarge · 48c · 384 GB RAM · 192 GB VRAM ───
 # Mistral Medium 3.5 128B là model dense ~80 GB, mạnh đồng thời ở instruction,
@@ -231,26 +231,26 @@ GRAPH_ENTITY_EXTRACTION=false
 # Gemma4:12b (~7.6 GB) dành riêng Designer và fit hoàn toàn trong 12 GB VRAM.
 # MAX_LOADED_MODELS=1 để không tranh VRAM. Nếu quá chậm, fallback hợp lý là
 # qwen3.5:9b cho reasoning và coding; nếu OOM, giảm CONTEXT_LENGTH xuống 8192.
-OLLAMA_MAX_LOADED_MODELS=1
-OLLAMA_NUM_PARALLEL=1
-OLLAMA_CONTEXT_LENGTH=16384
-OLLAMA_REQUEST_TIMEOUT=2400
-RAG_TIMEOUT=1800
-CHAT_MODEL=qwen3.6:27b
-EMBEDDING_MODEL=qwen3-embedding:8b
-CODING_PLANNER_MODEL=granite4.1:8b
-BA_MODEL=qwen3.6:27b
-PM_MODEL=qwen3.6:27b
-SA_MODEL=qwen3.6:27b
-TA_MODEL=qwen3.6:27b
-DESIGNER_MODEL=gemma4:12b
-FE_MODEL=north-mini-code-1.0:q4_K_M
-MOBILE_MODEL=north-mini-code-1.0:q4_K_M
-BE_MODEL=north-mini-code-1.0:q4_K_M
-DBA_MODEL=north-mini-code-1.0:q4_K_M
-DA_MODEL=qwen3.6:27b
-TL_MODEL=north-mini-code-1.0:q4_K_M
-TECH_LEAD_MODEL=north-mini-code-1.0:q4_K_M
-TESTER_MODEL=qwen3.6:27b
-DEVSECOPS_MODEL=north-mini-code-1.0:q4_K_M
-CLARIFIER_MODEL=qwen3.6:27b
+# OLLAMA_MAX_LOADED_MODELS=1
+# OLLAMA_NUM_PARALLEL=1
+# OLLAMA_CONTEXT_LENGTH=16384
+# OLLAMA_REQUEST_TIMEOUT=2400
+# RAG_TIMEOUT=1800
+# CHAT_MODEL=qwen3.6:27b
+# EMBEDDING_MODEL=qwen3-embedding:8b
+# CODING_PLANNER_MODEL=granite4.1:8b
+# BA_MODEL=qwen3.6:27b
+# PM_MODEL=qwen3.6:27b
+# SA_MODEL=qwen3.6:27b
+# TA_MODEL=qwen3.6:27b
+# DESIGNER_MODEL=gemma4:12b
+# FE_MODEL=north-mini-code-1.0:q4_K_M
+# MOBILE_MODEL=north-mini-code-1.0:q4_K_M
+# BE_MODEL=north-mini-code-1.0:q4_K_M
+# DBA_MODEL=north-mini-code-1.0:q4_K_M
+# DA_MODEL=qwen3.6:27b
+# TL_MODEL=north-mini-code-1.0:q4_K_M
+# TECH_LEAD_MODEL=north-mini-code-1.0:q4_K_M
+# TESTER_MODEL=qwen3.6:27b
+# DEVSECOPS_MODEL=north-mini-code-1.0:q4_K_M
+# CLARIFIER_MODEL=qwen3.6:27b