ltcai 2.1.0 → 2.2.0

package/llm_router.py

@@ -29,16 +29,14 @@ executor = ThreadPoolExecutor(max_workers=1)
 
 try:
     import mlx.core as mx
-    from mlx_lm import load as lm_load
     from mlx_vlm import load as vlm_load
     VLM_AVAILABLE = True
-    print("✅ MLX-VLM and MLX-LM are ready for Gemma 4.")
+    print("✅ MLX-VLM is ready for multimodal models.")
 except Exception as e:
     mx = None
-    lm_load = None
     vlm_load = None
     VLM_AVAILABLE = False
-    print(f"⚠️ MLX libraries unavailable: {e}")
+    print(f"⚠️ MLX-VLM unavailable: {e}")
 
 BRAND_NAME = "Lattice AI"
 LEGACY_BRAND_PATTERNS = [
@@ -77,12 +75,12 @@ OPENAI_COMPATIBLE_PROVIDERS = {
     "groq": {
         "env_key": "GROQ_API_KEY",
         "base_url": "https://api.groq.com/openai/v1",
-        "default_model": "llama-3.1-8b-instant",
+        "default_model": "meta-llama/llama-4-scout-17b-16e-instruct",
     },
     "together": {
         "env_key": "TOGETHER_API_KEY",
         "base_url": "https://api.together.xyz/v1",
-        "default_model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+        "default_model": "Qwen/Qwen3-VL-32B-Instruct",
     },
     "xai": {
         "env_key": "XAI_API_KEY",
@@ -93,14 +91,14 @@ OPENAI_COMPATIBLE_PROVIDERS = {
         "env_key": "OLLAMA_API_KEY",
         "base_url_env": "OLLAMA_BASE_URL",
         "base_url": "http://localhost:11434/v1",
-        "default_model": "llama3.1",
+        "default_model": "hf.co/ggml-org/gemma-4-12B-it-GGUF:Q4_K_M",
         "api_key_fallback": "ollama",
     },
     "vllm": {
         "env_key": "VLLM_API_KEY",
         "base_url_env": "VLLM_BASE_URL",
         "base_url": "http://localhost:8000/v1",
-        "default_model": "meta-llama/Llama-3.1-8B-Instruct",
+        "default_model": "Qwen/Qwen3-VL-8B-Instruct",
         "api_key_fallback": "vllm",
     },
     "lmstudio": {
@@ -137,20 +135,18 @@ PROVIDER_MODEL_CATALOG = {
         {"id": "anthropic/claude-sonnet-4.6", "name": "Claude Sonnet 4.6 via OpenRouter", "family": "Claude"},
         {"id": "anthropic/claude-haiku-4.5", "name": "Claude Haiku 4.5 via OpenRouter", "family": "Claude"},
         {"id": "qwen/qwen3-vl-235b-a22b-instruct", "name": "Qwen3-VL 235B A22B via OpenRouter", "family": "Qwen"},
-        {"id": "qwen/qwen3-coder", "name": "Qwen3 Coder via OpenRouter", "family": "Qwen"},
+        {"id": "google/gemma-4-12b-it", "name": "Gemma 4 12B via OpenRouter", "family": "Gemma"},
         {"id": "x-ai/grok-2", "name": "Grok 2 via OpenRouter", "family": "Grok"},
-        {"id": "meta-llama/llama-3.3-70b-instruct", "name": "Llama 3.3 70B via OpenRouter", "family": "Llama"},
+        {"id": "meta-llama/llama-4-scout-17b-16e-instruct", "name": "Llama 4 Scout via OpenRouter", "family": "Llama"},
         {"id": "google/gemini-2.5-flash", "name": "Gemini 2.5 Flash via OpenRouter", "family": "Gemini"},
     ],
     "groq": [
-        {"id": "qwen/qwen3-32b", "name": "Qwen3 32B", "family": "Qwen"},
-        {"id": "llama-3.1-8b-instant", "name": "Llama 3.1 8B Instant", "family": "Llama"},
-        {"id": "llama-3.3-70b-versatile", "name": "Llama 3.3 70B Versatile", "family": "Llama"},
+        {"id": "meta-llama/llama-4-scout-17b-16e-instruct", "name": "Llama 4 Scout", "family": "Llama"},
     ],
     "together": [
         {"id": "Qwen/Qwen3-VL-32B-Instruct", "name": "Qwen3-VL 32B", "family": "Qwen"},
-        {"id": "meta-llama/Llama-3.3-70B-Instruct-Turbo", "name": "Llama 3.3 70B Turbo", "family": "Llama"},
-        {"id": "mistralai/Mixtral-8x22B-Instruct-v0.1", "name": "Mixtral 8x22B", "family": "Mistral"},
+        {"id": "google/gemma-4-12b-it", "name": "Gemma 4 12B", "family": "Gemma"},
+        {"id": "meta-llama/Llama-4-Scout-17B-16E-Instruct", "name": "Llama 4 Scout", "family": "Llama"},
     ],
     "xai": [
         {"id": "grok-beta", "name": "Grok Beta", "family": "Grok"},
@@ -158,6 +154,40 @@ PROVIDER_MODEL_CATALOG = {
     ],
 }
 
+MODEL_SOURCE_BY_FAMILY = {
+    "GPT": ("미국", "OpenAI"),
+    "Claude": ("미국", "Anthropic"),
+    "Qwen": ("중국", "Alibaba"),
+    "Llama": ("미국", "Meta"),
+    "Gemini": ("미국", "Google"),
+    "Grok": ("미국", "xAI"),
+}
+
+
+def source_metadata_for_model(provider: str, model: Dict[str, str], *, local_server: bool) -> Dict[str, str]:
+    family = str(model.get("family") or "")
+    country, company = MODEL_SOURCE_BY_FAMILY.get(family, ("미상", provider.title()))
+    if local_server:
+        execution_method = "내 컴퓨터에서만 실행"
+        internet_requirement = "모델을 다운로드할 때만 인터넷 필요; 실행 중에는 필요 없음"
+    else:
+        execution_method = "인터넷 연결 후 사용"
+        internet_requirement = "내 파일이 인터넷으로 전송될 수 있음"
+    return {
+        "source_country": country,
+        "source_company": company,
+        "execution_method": execution_method,
+        "internet_requirement": internet_requirement,
+        "model_name": model.get("name") or model.get("id") or "",
+        "source_display_order": [
+            "source_country",
+            "source_company",
+            "execution_method",
+            "internet_requirement",
+            "model_name",
+        ],
+    }
+
 @dataclass
 class CloudModel:
     provider: str
@@ -207,37 +237,29 @@ def _resolve_local_hf_model(model_id: str) -> str:
     return model_id
 
 def ensure_mlx_runtime() -> None:
-    global mx, lm_load, vlm_load, VLM_AVAILABLE
-    if mx is not None and lm_load is not None:
+    global mx, vlm_load, VLM_AVAILABLE
+    if mx is not None and vlm_load is not None:
         return
     try:
         import mlx.core as mlx_core
-        from mlx_lm import load as mlx_lm_load
+        from mlx_vlm import load as mlx_vlm_load
 
         mx = mlx_core
-        lm_load = mlx_lm_load
-        try:
-            from mlx_vlm import load as mlx_vlm_load
-            vlm_load = mlx_vlm_load
-            VLM_AVAILABLE = True
-        except Exception:
-            vlm_load = None
-            VLM_AVAILABLE = False
+        vlm_load = mlx_vlm_load
+        VLM_AVAILABLE = True
         mx.set_default_device(mx.gpu)
     except Exception as e:
-        raise RuntimeError(f"MLX runtime is not available after install: {e}") from e
+        raise RuntimeError(f"MLX-VLM runtime is not available after install: {e}") from e
 
 def _mlx_sampler(temperature: float):
     """Build an MLX sampler callable for the given temperature.
 
-    mlx_lm >= 0.20 removed the ``temp`` keyword from generate_step in favour of a
-    ``sampler`` callable, and mlx_vlm follows the same convention. Passing
-    ``temp=`` to generate/stream_generate now raises
-    ``generate_step() got an unexpected keyword argument 'temp'``. Both libraries
-    accept ``sampler=`` and share make_sampler from mlx_lm.sample_utils.
+    Lattice v2.2 keeps local execution on MLX-VLM only. Returning ``None`` lets
+    MLX-VLM use its bundled default sampler without pulling another generation
+    package into the runtime contract.
     """
-    from mlx_lm.sample_utils import make_sampler
-    return make_sampler(temp=temperature)
+    _ = temperature
+    return None
 
 class LLMRouter:
     def __init__(self):
@@ -331,8 +353,8 @@ class LLMRouter:
             return self._load_cloud_model(provider, provider_model, api_key_override=api_key_override, owner=owner)
 
         ensure_mlx_runtime()
-        if mx is None or lm_load is None:
-            raise RuntimeError("MLX is not available in this process. Run on Apple Silicon with Metal access.")
+        if mx is None or vlm_load is None:
+            raise RuntimeError("MLX-VLM is not available in this process. Run on Apple Silicon with Metal access.")
 
         cache_key = f"{model_id}_{draft_model_id}" if draft_model_id else model_id
         if cache_key in self._cache:
@@ -348,24 +370,13 @@ class LLMRouter:
         
         def _load():
             mx.set_default_device(mx.gpu)
-            is_gemma4 = "gemma-4" in model_id.lower() or "gemma4" in model_id.lower()
-            
-            # 1. Target 로드 (Gemma 4는 항상 vlm_load 사용)
-            if is_gemma4 and VLM_AVAILABLE:
-                print(f"🔄 Loading Target (VLM Mode): {target_model_id}...")
-                model, tokenizer = vlm_load(target_model_id)
-            else:
-                print(f"🔄 Loading Target (LM Mode): {target_model_id}...")
-                model, tokenizer = lm_load(target_model_id)
-
-            # 2. Draft 로드 (Gemma 4는 항상 vlm_load 사용)
+            print(f"🔄 Loading Target (VLM Mode): {target_model_id}...")
+            model, tokenizer = vlm_load(target_model_id)
+
             draft_model = None
             if target_draft_model_id:
                 print(f"🔄 Loading Assistant (VLM Mode): {target_draft_model_id}...")
-                if is_gemma4 and VLM_AVAILABLE:
-                    draft_model, _ = vlm_load(target_draft_model_id)
-                else:
-                    draft_model, _ = lm_load(target_draft_model_id)
+                draft_model, _ = vlm_load(target_draft_model_id)
                 print(f"✅ Assistant Ready.")
 
             return model, tokenizer, draft_model
@@ -418,14 +429,16 @@ class LLMRouter:
             }]
             for model in provider_models:
                 model_id = model["id"]
+                local_server = provider in local_server_providers
                 items.append({
                     "id": f"{provider}:{model_id}",
                     "name": model.get("name") or f"{provider.title()} · {model_id}",
                     "provider": provider,
                     "family": model.get("family"),
-                    "tag": "local-server" if provider in local_server_providers else "cloud",
+                    "tag": "local-server" if local_server else "cloud",
                     "available": has_key,
                     "requires": config["env_key"] if not has_key else None,
+                    **source_metadata_for_model(provider, model, local_server=local_server),
                 })
         custom = os.getenv("LATTICEAI_CLOUD_MODELS") or ""
         for raw in [item.strip() for item in custom.split(",") if item.strip()]:
@@ -439,6 +452,11 @@ class LLMRouter:
                     "tag": "cloud",
                     "available": bool(os.getenv(config["env_key"]) or config.get("api_key_fallback")),
                     "requires": None,
+                    **source_metadata_for_model(
+                        provider,
+                        {"id": model, "name": f"{provider.title()} · {model}", "family": provider.title()},
+                        local_server=provider in local_server_providers,
+                    ),
                 })
         return items
 
@@ -511,25 +529,15 @@ class LLMRouter:
             return await self._cloud_generate(cached, message, context, max_tokens, temperature)
 
         model, tokenizer, draft_model = self._cache[self._current]
-        is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-        prompt = (
-            self._build_vlm_prompt(model, tokenizer, message, context, 1)
-            if image_data and is_gemma4 and VLM_AVAILABLE
-            else self._build_prompt(message, context, tokenizer)
-        )
+        prompt = self._build_vlm_prompt(model, tokenizer, message, context, 1 if image_data else 0)
         
         loop = asyncio.get_event_loop()
         
         def _gen():
             import mlx.core as mx
             mx.set_default_device(mx.gpu)
-            is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-            if is_gemma4 and VLM_AVAILABLE:
-                from mlx_vlm import generate as vlm_gen
-                return vlm_gen(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
-            else:
-                from mlx_lm import generate as lm_gen
-                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
+            from mlx_vlm import generate as vlm_gen
+            return vlm_gen(model, tokenizer, prompt=prompt, image=self._prep_image(image_data) if image_data else None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
         result = await loop.run_in_executor(executor, _gen)
         # mlx-vlm might return a GenerationResult object; extract the text
         if hasattr(result, "text"):
@@ -567,12 +575,7 @@ class LLMRouter:
             return
 
         model, tokenizer, draft_model = self._cache[self._current]
-        is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-        prompt = (
-            self._build_vlm_prompt(model, tokenizer, message, context, 1)
-            if image_data and is_gemma4 and VLM_AVAILABLE
-            else self._build_prompt(message, context, tokenizer)
-        )
+        prompt = self._build_vlm_prompt(model, tokenizer, message, context, 1 if image_data else 0)
         loop = asyncio.get_event_loop()
         queue = asyncio.Queue()
 
@@ -580,13 +583,8 @@ class LLMRouter:
             import mlx.core as mx
             mx.set_default_device(mx.gpu)
             try:
-                is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-                if is_gemma4 and VLM_AVAILABLE:
-                    from mlx_vlm import stream_generate as vlm_stream
-                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
-                else:
-                    from mlx_lm import stream_generate as lm_stream
-                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
+                from mlx_vlm import stream_generate as vlm_stream
+                gen = vlm_stream(model, tokenizer, prompt=prompt, image=self._prep_image(image_data) if image_data else None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
                 
                 for chunk in gen:
                     text = chunk.text if hasattr(chunk, "text") else (chunk[0] if isinstance(chunk, tuple) else str(chunk))
@@ -675,13 +673,8 @@ class LLMRouter:
         def _gen():
             import mlx.core as mx
             mx.set_default_device(mx.gpu)
-            is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-            if is_gemma4 and VLM_AVAILABLE:
-                from mlx_vlm import generate as vlm_gen
-                return vlm_gen(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
-            else:
-                from mlx_lm import generate as lm_gen
-                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
+            from mlx_vlm import generate as vlm_gen
+            return vlm_gen(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
         result = await loop.run_in_executor(executor, _gen)
         if hasattr(result, "text"):
             return normalize_branding(result.text)
@@ -742,13 +735,8 @@ class LLMRouter:
             import mlx.core as mx
             mx.set_default_device(mx.gpu)
             try:
-                is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
-                if is_gemma4 and VLM_AVAILABLE:
-                    from mlx_vlm import stream_generate as vlm_stream
-                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
-                else:
-                    from mlx_lm import stream_generate as lm_stream
-                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
+                from mlx_vlm import stream_generate as vlm_stream
+                gen = vlm_stream(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
                 for chunk in gen:
                     text = chunk.text if hasattr(chunk, "text") else (chunk[0] if isinstance(chunk, tuple) else str(chunk))
                     loop.call_soon_threadsafe(queue.put_nowait, text)

package/ltcai_cli.py

@@ -97,9 +97,8 @@ def doctor() -> int:
         ("FastAPI", _has_module("fastapi"), "required server dependency", True),
         ("Uvicorn", _has_module("uvicorn"), "required server dependency", True),
         ("OpenAI SDK", _has_module("openai"), "required for cloud providers", False),
-        ("MLX", _has_module("mlx"), "required for Apple Silicon local models", False),
-        ("MLX-LM", _has_module("mlx_lm"), "required for local text models", False),
-        ("MLX-VLM", _has_module("mlx_vlm"), "required for Gemma/VLM models", False),
+        ("MLX", _has_module("mlx"), "required for Apple Silicon multimodal models", False),
+        ("MLX-VLM", _has_module("mlx_vlm"), "required for Gemma-4/VLM models", False),
         ("Ollama binary", shutil.which("ollama") is not None, "optional local-server engine", False),
     ]
     data_dir = Path(os.getenv("LATTICEAI_DATA_DIR") or Path.home() / ".ltcai")

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ltcai",
-  "version": "2.1.0",
+  "version": "2.2.0",
   "description": "Lattice AI Workspace OS for local-first graph, memory, agent, workflow, and skill operations",
   "homepage": "https://github.com/TaeSooPark-PTS/LatticeAI#readme",
   "repository": {

package/static/chat.html

@@ -237,8 +237,7 @@
                                 <button class="hdc-btn" onclick="openDataGraph()"><i class="ti ti-arrow-right"></i> 그래프 보기</button>
                             </div>
 
-                            <!-- 자동 설정 (고급/관리자 모드만) -->
-                            <div class="hdc-card hdc-setup" id="home-setup-card" style="display:none">
+                            <div class="hdc-card hdc-setup" id="home-setup-card">
                                 <div class="hdc-title"><i class="ti ti-settings-automation"></i> 자동 설정</div>
                                 <div class="hdc-setup-count">
                                     <span id="home-setup-num">—</span>
@@ -279,7 +278,7 @@
                         <div class="chat-capability-row" id="chat-capability-row">
                             <span>파일 생성</span>
                             <span>지식 정리</span>
-                            <span>로컬 런타임</span>
+                            <span>내 컴퓨터에서 실행</span>
                         </div>
                     </div>
                 </section>
@@ -359,12 +358,12 @@
                 <button class="mode-card" id="mode-card-advanced" onclick="selectMode('advanced')">
                     <div class="mode-icon"><i class="ti ti-terminal-2"></i></div>
                     <h3 data-i18n="mode_advanced">고급 모드</h3>
-                    <span data-i18n="mode_advanced_sub">모델 상태, 런타임 설정, 고급 도구 관리</span>
+                    <span data-i18n="mode_advanced_sub">같은 기능을 더 자세한 설명으로 표시</span>
                 </button>
                 <button class="mode-card" id="mode-card-admin" onclick="selectMode('admin')">
                     <div class="mode-icon"><i class="ti ti-shield-lock"></i></div>
                     <h3 data-i18n="mode_admin">관리자 모드</h3>
-                    <span data-i18n="mode_admin_sub">운영자용 관리자 대시보드</span>
+                    <span data-i18n="mode_admin_sub">사용자, 정책, 감사 로그 관리</span>
                 </button>
             </div>
         </section>
@@ -375,7 +374,7 @@
             <div class="model-panel-header">
                 <div>
                     <h2 data-i18n="model_switcher">모델 스위처</h2>
-                    <p style="color: var(--muted); font-size: 12px; margin-top: 4px;" data-i18n="model_switcher_sub">실행 엔진을 설치하고, 엔진에 맞는 local/cloud LLM을 선택합니다.</p>
+                    <p style="color: var(--muted); font-size: 12px; margin-top: 4px;" data-i18n="model_switcher_sub">제작 국가, 제작 회사, 실행 방식, 인터넷 사용 여부를 확인하고 모델을 선택합니다.</p>
                 </div>
                 <button class="admin-close" onclick="closeModelPanel()"><i class="ti ti-x"></i></button>
             </div>

package/static/scripts/chat.js

@@ -227,7 +227,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 my_status: '내 상태 보기', auto_setup: '자동 설정',
                 nav_home: '홈', nav_chat: '채팅', nav_workspace: 'Workspace OS', nav_knowledge: '지식 그래프',
                 nav_pipeline: '파이프라인', nav_files: '내 컴퓨터',
-                nav_model_status: '모델 상태', nav_runtime: '런타임 설정',
+                nav_model_status: '모델 상태', nav_runtime: '실행 방식 설정',
                 nav_advanced_settings: '고급 설정',
                 history_search_ph: '대화 검색...', new_chat: 'New Chat',
                 history_section: '대화', history_empty: '아직 저장된 대화가 없습니다.',
@@ -235,7 +235,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 confirm_delete_chat: '이 대화를 삭제할까요?',
                 home_greeting: '안녕하세요, {name}님',
                 home_greeting_short: '안녕하세요',
-                ops_ai_model: 'AI 모델', ops_local_runtime: '로컬 런타임',
+                ops_ai_model: 'AI 모델', ops_local_runtime: '내 컴퓨터에서 실행',
                 ops_admin_network: '관리자 네트워크', ops_admin_security: '관리자 보안',
                 ops_pipeline_value: '멀티 LLM 파이프라인',
                 ops_pipeline_meta: 'Plan → Execute → Review 모델 설정',
@@ -248,7 +248,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 home_recent_files: '최근 파일', home_open_files: '파일 열기', home_no_files: '파일이 없습니다',
                 chat_intro_title: 'Lattice AI',
                 chat_intro_desc: '로컬 모델, 파일, 지식 그래프, 멀티모달 작업을 한 대화 흐름에서 연결하는 개인 AI 워크스페이스입니다.',
-                chat_cap_file: '파일 생성', chat_cap_knowledge: '지식 정리', chat_cap_runtime: '로컬 런타임',
+                chat_cap_file: '파일 생성', chat_cap_knowledge: '지식 정리', chat_cap_runtime: '내 컴퓨터에서 실행',
                 // 계정 모달
                 tab_profile: '프로필', tab_password: '비밀번호',
                 label_name: '이름', label_nickname: '닉네임',
@@ -283,12 +283,12 @@ const chatViewport = document.getElementById('chat-viewport');
                 mode_default: '기본 모드',
                 mode_default_sub: '대화, 파일 생성, 지식 정리를 한 화면에서',
                 mode_advanced: '고급 모드',
-                mode_advanced_sub: '모델 상태, 런타임 설정, 고급 설정',
+                mode_advanced_sub: '같은 기능을 더 자세한 설명으로 표시',
                 mode_admin: '관리자 모드',
-                mode_admin_sub: '운영자용 관리자 대시보드',
+                mode_admin_sub: '사용자, 정책, 감사 로그 관리',
                 // 패널 제목
                 model_switcher: '모델 스위처',
-                model_switcher_sub: '실행 엔진을 설치하고, 엔진에 맞는 local/cloud LLM을 선택합니다.',
+                model_switcher_sub: '제작 국가, 제작 회사, 실행 방식, 인터넷 사용 여부를 확인하고 모델을 선택합니다.',
                 // 권한 다이얼로그
                 perm_title: '파일 접근 요청', btn_deny: '거부', btn_allow: '허용',
             },
@@ -306,7 +306,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 my_status: 'My Status', auto_setup: 'Auto Setup',
                 nav_home: 'Home', nav_chat: 'Chat', nav_workspace: 'Workspace OS', nav_knowledge: 'Knowledge Graph',
                 nav_pipeline: 'Pipeline', nav_files: 'My Computer',
-                nav_model_status: 'Model Status', nav_runtime: 'Runtime Settings',
+                nav_model_status: 'Model Status', nav_runtime: 'Execution Settings',
                 nav_advanced_settings: 'Advanced Settings',
                 history_search_ph: 'Search chats...', new_chat: 'New Chat',
                 history_section: 'Chats', history_empty: 'No saved chats yet.',
@@ -314,7 +314,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 confirm_delete_chat: 'Delete this chat?',
                 home_greeting: 'Hello, {name}',
                 home_greeting_short: 'Hello',
-                ops_ai_model: 'AI model', ops_local_runtime: 'Local runtime',
+                ops_ai_model: 'AI model', ops_local_runtime: 'Runs on this computer',
                 ops_admin_network: 'Admin Network', ops_admin_security: 'Admin Security',
                 ops_pipeline_value: 'Multi-LLM Pipeline',
                 ops_pipeline_meta: 'Plan → Execute → Review model setup',
@@ -327,7 +327,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 home_recent_files: 'Recent Files', home_open_files: 'Open Files', home_no_files: 'No files yet',
                 chat_intro_title: 'Lattice AI',
                 chat_intro_desc: 'A personal AI workspace that connects local models, files, knowledge graphs, and multimodal work in one conversation flow.',
-                chat_cap_file: 'File creation', chat_cap_knowledge: 'Knowledge organizing', chat_cap_runtime: 'Local runtime',
+                chat_cap_file: 'File creation', chat_cap_knowledge: 'Knowledge organizing', chat_cap_runtime: 'Runs on this computer',
                 // Account modal
                 tab_profile: 'Profile', tab_password: 'Password',
                 label_name: 'Name', label_nickname: 'Nickname',
@@ -367,7 +367,7 @@ const chatViewport = document.getElementById('chat-viewport');
                 mode_admin_sub: 'Admin dashboard for operators',
                 // Panel titles
                 model_switcher: 'Model Switcher',
-                model_switcher_sub: 'Install a runtime engine and select a local/cloud LLM.',
+                model_switcher_sub: 'Check maker country, maker company, execution method, internet use, then select a model.',
                 // Permission dialog
                 perm_title: 'File Access Request', btn_deny: 'Deny', btn_allow: 'Allow',
             }
@@ -650,11 +650,8 @@ const chatViewport = document.getElementById('chat-viewport');
         }
 
         async function _loadHomeDashboard() {
-            const mode = getCurrentMode();
-
-            // 자동 설정 카드: 고급/관리자 모드만
             const setupCard = document.getElementById('home-setup-card');
-            if (setupCard) setupCard.style.display = (mode === 'advanced' || mode === 'admin') ? 'flex' : 'none';
+            if (setupCard) setupCard.style.display = 'flex';
 
             // 모델 + sysinfo 병렬 fetch
             try {
@@ -953,7 +950,7 @@ const chatViewport = document.getElementById('chat-viewport');
             const selected = models.find(item => item.checked && !item.disabled && (item.model_id || item.action?.model_id))
                 || models.find(item => !item.disabled && (item.model_id || item.action?.model_id));
             const zero = onboardingRecs?.summary?.zero_config || onboardingEnv?.zero_config?.recommend || {};
-            const modelId = selected?.model_id || selected?.action?.model_id || zero.model_id || 'mlx-community/Llama-3.2-3B-Instruct-4bit';
+            const modelId = selected?.model_id || selected?.action?.model_id || zero.model_id || 'mlx-community/gemma-4-12b-it-4bit';
             const engineItem = (onboardingRecs?.engines || []).find(item => item.checked && !item.disabled);
             const runtime = engineItem?.name || (zero.runtime === 'mlx' ? 'MLX' : zero.runtime) || 'MLX';
             return {
@@ -1351,7 +1348,7 @@ const chatViewport = document.getElementById('chat-viewport');
                     <button class="onboarding-mode" onclick="finishOnboarding('advanced')">
                         <i class="ti ti-terminal-2"></i>
                         <h3>고급 모드</h3>
-                        <p>모델 상태, 런타임 설정, 고급 설정까지 함께 다룹니다.</p>
+                        <p>같은 기능을 유지하면서 모델, 메모리, 실행 방식 설명을 더 자세히 표시합니다.</p>
                     </button>
                     ${adminCard}
                 </div>
@@ -1528,7 +1525,7 @@ const chatViewport = document.getElementById('chat-viewport');
             const isUnavailable = unsupported || (!isLocalEngine && engineMissing) || keyMissing || verifyFailed;
             const badge = unsupported ? '현재 환경 미지원'
                 : engineMissing && isLocalEngine ? '설치 후 자동 로드'
-                : engineMissing ? '엔진 설치 필요'
+                : engineMissing ? '실행 도구 설치 필요'
                 : needsPull ? '다운로드 후 자동 로드'
                 : keyMissing ? `필요: ${model.requires || 'API key'}`
                 : verifyFailed ? `실패: ${model.verify_reason || '검증 실패'}`
@@ -1539,11 +1536,19 @@ const chatViewport = document.getElementById('chat-viewport');
             const action = isLocalEngine
                 ? `selectModelByCard('${encodeURIComponent(model.id)}', '${engine?.id || ''}')`
                 : `loadSelectedModel('${encodeURIComponent(model.id)}', '${engine?.id || ''}')`;
+            const sourceLine = [
+                model.source_country,
+                model.source_company,
+                model.execution_method,
+                model.internet_requirement,
+                model.model_name || model.name,
+            ].filter(Boolean).join(' · ');
+            const detailLine = sourceLine || `${model.id} · ${badge}`;
             return `
                 <button class="model-option${cls}" ${isUnavailable ? 'disabled' : ''} onclick="${action}">
                     <div>
                         <strong>${escapeHtml(model.name || compactModelName(model.id))}</strong>
-                        <span>${escapeHtml(model.id)} · ${escapeHtml(badge)}</span>
+                        <span>${escapeHtml(detailLine)}${sourceLine ? `<br>${escapeHtml(model.id)} · ${escapeHtml(badge)}` : ''}</span>
                     </div>
                     <i class="ti ${icon}"></i>
                 </button>
@@ -1556,12 +1561,9 @@ const chatViewport = document.getElementById('chat-viewport');
             if (raw.includes('claude')) return 'Claude';
             if (raw.includes('grok')) return 'Grok';
             if (raw.includes('gemini')) return 'Gemini';
-            if (raw.includes('mistral') || raw.includes('mixtral')) return 'Mistral';
             if (raw.includes('qwen')) return 'Qwen';
             if (raw.includes('llama')) return 'Llama';
             if (raw.includes('gemma')) return 'Gemma';
-            if (raw.includes('phi')) return 'Phi';
-            if (raw.includes('deepseek')) return 'DeepSeek';
             return (model?.family || '기타');
         }
 
@@ -1642,17 +1644,17 @@ const chatViewport = document.getElementById('chat-viewport');
             const cloudEngines = cachedEngineList.filter(engine => engine.kind === 'cloud');
             const isLocal = modelPanelFilter === 'local';
             const target = isLocal ? localEngines : cloudEngines;
-            const emptyText = isLocal ? '등록된 로컬 엔진이 없습니다.' : '등록된 클라우드 엔진이 없습니다.';
+            const emptyText = isLocal ? '내 컴퓨터에서 실행할 수 있는 항목이 없습니다.' : '인터넷 연결 후 사용할 수 있는 항목이 없습니다.';
 
             modelList.innerHTML = `
-                <div class="model-group-title">EXECUTION ENGINES</div>
+                <div class="model-group-title">실행 방식</div>
                 <div class="model-filter">
-                    <button class="model-filter-btn ${isLocal ? 'active' : ''}" onclick="setModelPanelFilter('local')">Local LLM</button>
-                    <button class="model-filter-btn ${!isLocal ? 'active' : ''}" onclick="setModelPanelFilter('cloud')">Cloud LLM</button>
+                    <button class="model-filter-btn ${isLocal ? 'active' : ''}" onclick="setModelPanelFilter('local')">내 컴퓨터에서만 실행</button>
+                    <button class="model-filter-btn ${!isLocal ? 'active' : ''}" onclick="setModelPanelFilter('cloud')">인터넷 연결 후 사용</button>
                 </div>
                 ${!isLocal ? `
                     <div style="display:flex;justify-content:flex-end;margin:-2px 0 8px;">
-                        <button class="admin-action" onclick="verifyCloudModels(true)"><i class="ti ti-activity"></i> Cloud 실사용 테스트</button>
+                        <button class="admin-action" onclick="verifyCloudModels(true)"><i class="ti ti-activity"></i> 인터넷 모델 실사용 테스트</button>
                     </div>
                 ` : ''}
                 ${target.length ? target.map(engineCardHtml).join('') : `<div class="sensitivity-preview">${emptyText}</div>`}
@@ -1680,7 +1682,7 @@ const chatViewport = document.getElementById('chat-viewport');
 
         async function verifyCloudModels(force = true) {
             const modelList = document.getElementById('model-list');
-            modelList.innerHTML = `<div class="sensitivity-preview">Cloud 모델 실사용 테스트 중입니다... (provider별로 수 초~수십 초)</div>`;
+            modelList.innerHTML = `<div class="sensitivity-preview">인터넷 모델 실사용 테스트 중입니다... (연결 방식별로 수 초~수십 초)</div>`;
             try {
                 const res = await apiFetch('/engines/verify-cloud', {
                     method: 'POST',
@@ -1688,9 +1690,9 @@ const chatViewport = document.getElementById('chat-viewport');
                     body: JSON.stringify({ force })
                 });
                 const data = await res.json();
-                if (!res.ok) throw new Error(data.detail || 'Cloud 실사용 테스트 실패');
+                if (!res.ok) throw new Error(data.detail || '인터넷 모델 실사용 테스트 실패');
                 await openModelPanel();
-                addMessage('ai', `Cloud 모델 실사용 테스트를 완료했습니다. 실패한 모델은 잠금 상태로 표시됩니다.`);
+                addMessage('ai', `인터넷 모델 실사용 테스트를 완료했습니다. 실패한 모델은 잠금 상태로 표시됩니다.`);
             } catch (e) {
                 modelList.innerHTML = `
                     <div class="sensitivity-preview">${escapeHtml(e.message)}</div>
@@ -1832,7 +1834,7 @@ const chatViewport = document.getElementById('chat-viewport');
                         </div>
                     </div>
                     <div id="model-download-detail" class="model-download-detail">
-                        엔진 설치, 모델 다운로드, 서버 시작, 로드까지 자동으로 진행합니다. 첫 실행은 수 분이 걸릴 수 있습니다.
+                        실행 도구 설치, 모델 다운로드, 연결 준비, 로드까지 자동으로 진행합니다. 첫 실행은 수 분이 걸릴 수 있습니다.
                     </div>
                 </div>
             `;
@@ -4363,7 +4365,7 @@ const chatViewport = document.getElementById('chat-viewport');
         const keys = env.api_keys || {};
 
         const mlxLabel = mlx.available
-            ? (mlx.mlx_lm && mlx.mlx_vlm ? 'MLX-LM · MLX-VLM 설치됨' : mlx.mlx_lm ? 'MLX-LM 설치됨' : '부분 설치')
+            ? (mlx.mlx_vlm ? 'MLX-VLM 설치됨' : 'MLX 설치됨 · MLX-VLM 필요')
             : '미설치';
 
         const cloudKeys = Object.entries(keys).filter(([,v]) => v).map(([k]) => k.toUpperCase());
@@ -4378,7 +4380,7 @@ const chatViewport = document.getElementById('chat-viewport');
             { icon: mlx.available ? '✅' : '⚠️', label: 'MLX', value: mlxLabel, ok: mlx.available },
             { icon: tools.ollama ? '✅' : '○',  label: 'Ollama', value: tools.ollama ? '설치됨' : '미설치', ok: true },
             { icon: tools.brew   ? '✅' : '○',  label: 'Homebrew', value: tools.brew ? '설치됨' : '미설치', ok: true },
-            { icon: cloudKeys.length ? '✅' : '○', label: 'Cloud API',
+            { icon: cloudKeys.length ? '✅' : '○', label: '인터넷 AI',
               value: cloudKeys.length ? cloudKeys.join(', ') : '없음', ok: true },
             { icon: env.os === 'Darwin' ? '🍎' : '🐧',
               label: '운영체제',
@@ -4607,10 +4609,6 @@ const chatViewport = document.getElementById('chat-viewport');
     let _mcpCurrentTab = 'registry';
 
     async function openMcpModal() {
-        if (getCurrentMode() === 'default') {
-            showToast('고급 모드에서 사용할 수 있습니다.');
-            return;
-        }
         document.getElementById('mcp-modal-overlay').classList.add('open');
         await renderMcpModal(_mcpCurrentTab);
     }