ltcai 2.0.0 → 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": "2.0.0",
+  "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": {
@@ -19,7 +19,7 @@
     "dev": "python3 ltcai_cli.py --reload",
     "build": "npm run build:python",
     "build:python": "python3 -m build",
-    "check:python": "python3 -m py_compile ltcai_cli.py server.py latticeai/server_app.py latticeai/api/chat.py latticeai/api/computer_use.py latticeai/api/deps.py latticeai/api/garden.py latticeai/api/local_files.py latticeai/api/permissions.py latticeai/api/setup.py latticeai/api/static_routes.py latticeai/api/tools.py latticeai/api/plugins.py latticeai/api/workflow_designer.py latticeai/api/agents.py latticeai/api/realtime.py latticeai/services/app_context.py latticeai/services/model_runtime.py latticeai/services/model_catalog.py latticeai/services/model_recommendation.py latticeai/services/tool_dispatch.py latticeai/services/upload_service.py latticeai/core/tool_registry.py latticeai/core/enterprise.py latticeai/core/enterprise_admin.py latticeai/core/agent_prompts.py latticeai/core/workspace_os.py latticeai/core/plugins.py latticeai/core/workflow_engine.py latticeai/core/multi_agent.py latticeai/core/realtime.py knowledge_graph.py knowledge_graph_api.py local_knowledge_api.py llm_router.py p_reinforce.py telegram_bot.py tools.py codex_telegram_bot.py",
+    "check:python": "python3 -m py_compile ltcai_cli.py server.py latticeai/server_app.py latticeai/api/chat.py latticeai/api/computer_use.py latticeai/api/deps.py latticeai/api/garden.py latticeai/api/local_files.py latticeai/api/permissions.py latticeai/api/setup.py latticeai/api/static_routes.py latticeai/api/tools.py latticeai/api/plugins.py latticeai/api/workflow_designer.py latticeai/api/agents.py latticeai/api/realtime.py latticeai/api/marketplace.py latticeai/services/app_context.py latticeai/services/model_runtime.py latticeai/services/model_catalog.py latticeai/services/model_recommendation.py latticeai/services/tool_dispatch.py latticeai/services/upload_service.py latticeai/core/tool_registry.py latticeai/core/enterprise.py latticeai/core/enterprise_admin.py latticeai/core/agent_prompts.py latticeai/core/workspace_os.py latticeai/core/plugins.py latticeai/core/marketplace.py latticeai/core/workflow_engine.py latticeai/core/multi_agent.py latticeai/core/realtime.py knowledge_graph.py knowledge_graph_api.py local_knowledge_api.py llm_router.py p_reinforce.py telegram_bot.py tools.py codex_telegram_bot.py",
     "test": "python3 -m pytest tests/ -v",
     "test:unit": "python3 -m pytest tests/unit/ -v",
     "test:integration": "python3 -m pytest tests/integration/ -v",

package/static/agents.html

@@ -9,7 +9,7 @@
 <body>
   <main>
     <h1>Multi-Agent Runtime</h1>
-    <p class="sub">Planner · Executor · Reviewer · Researcher · Release — with handoff, retry, and an observable timeline.</p>
+    <p class="sub">Planner · Executor · Reviewer · Researcher · Release — with handoff, context packets, memory, retry, and replay.</p>
 
     <div class="section">
       <label>Goal</label>
@@ -28,6 +28,11 @@
       <h3>Recent agent runs</h3>
       <div id="runs"><div class="empty">Loading…</div></div>
     </div>
+
+    <div class="section">
+      <h3>Replay viewer</h3>
+      <div id="replay"><div class="empty">Select a recent run.</div></div>
+    </div>
   </main>
 
   <script type="module">
@@ -47,13 +52,33 @@
 
     function renderTimeline(timeline) {
       return (timeline || []).map((t) => {
-        const label = t.event === "handoff" ? `↪ handoff ${escapeHtml(t.from)} → ${escapeHtml(t.to)}`
+        const label = (t.event || "").startsWith("handoff_") ? `↪ ${escapeHtml(t.event)} ${escapeHtml(t.from||"")} → ${escapeHtml(t.to||"")}`
+          : t.event === "handoff" ? `↪ handoff ${escapeHtml(t.from)} → ${escapeHtml(t.to)}`
           : t.event === "role" ? `● ${escapeHtml(t.role)} ${badge(t.status)}`
+          : t.event === "retry_requested" ? `↻ retry ${escapeHtml(t.reason||"")}`
+          : t.event === "review_approved" ? `✓ review approved`
           : `· ${escapeHtml(t.event)}`;
         return `<div class="timeline-item">${label}<div class="t-meta">${escapeHtml(t.note||t.timestamp||"")}</div></div>`;
       }).join("");
     }
 
+    function renderHandoffs(handoffs) {
+      if (!handoffs?.length) return `<div class="empty">No handoffs recorded.</div>`;
+      return handoffs.map((h) => `<div class="timeline-item">
+        <strong>${escapeHtml(h.handoff_id)}</strong> ${badge(h.status)}
+        <div class="t-meta">${escapeHtml(h.source_agent)} → ${escapeHtml(h.target_agent)} · ${escapeHtml(h.reason||"")}</div>
+      </div>`).join("");
+    }
+
+    function renderReview(result) {
+      const reviews = result.review_history || [];
+      const retries = result.retry_history || [];
+      return `<div class="grid two">
+        <div>${reviews.length ? reviews.map((r) => `<div class="timeline-item">${badge(r.outcome)} ${escapeHtml(r.reason||"")}<div class="t-meta">retry ${r.retry_count}</div></div>`).join("") : `<div class="empty">No review history.</div>`}</div>
+        <div>${retries.length ? retries.map((r) => `<div class="timeline-item">${badge("retry " + r.retry)} ${escapeHtml(r.reason||"")}<div class="t-meta">limit ${r.limit}</div></div>`).join("") : `<div class="empty">No retries.</div>`}</div>
+      </div>`;
+    }
+
     document.getElementById("runBtn").addEventListener("click", async () => {
       const btn = document.getElementById("runBtn");
       btn.disabled = true;
@@ -66,7 +91,9 @@
           <div class="card">
             <div class="row"><h3>${escapeHtml(r.output)}</h3><div class="spacer"></div>${badge(r.status)}</div>
             <div class="meta">retries: ${r.retries} · roles: ${(r.roles_run||[]).join(" → ")}</div>
-            <div class="section">${renderTimeline(r.timeline)}</div>
+            <div class="section"><h3>Handoff chain</h3>${renderHandoffs(r.handoffs)}</div>
+            <div class="section"><h3>Review panel</h3>${renderReview(r)}</div>
+            <div class="section"><h3>Timeline</h3>${renderTimeline(r.timeline)}</div>
           </div>`;
         toast(`Agent run: ${r.status}`);
         await loadRuns();
@@ -82,9 +109,26 @@
         <div class="card" style="margin-bottom:10px">
           <div class="row"><h3>${escapeHtml((r.input||"").slice(0,80))}</h3><div class="spacer"></div>${badge(r.status)}</div>
           <div class="meta">${escapeHtml(r.agent_id)} · ${escapeHtml(r.created_at)} · ${(r.timeline||[]).length} timeline events</div>
+          <div class="row" style="margin-top:10px"><button class="ghost" data-replay="${r.id}">Replay</button></div>
         </div>`).join("");
     }
 
+    document.getElementById("runs").addEventListener("click", async (e) => {
+      const btn = e.target.closest("button[data-replay]");
+      if (!btn) return;
+      const out = document.getElementById("replay");
+      out.innerHTML = `<div class="empty">Loading replay…</div>`;
+      try {
+        const data = await api(`/agents/api/runs/${btn.dataset.replay}/replay`);
+        const frames = data.replay.frames || [];
+        out.innerHTML = frames.map((f) => `<div class="timeline-item">
+          <div class="row"><strong>${escapeHtml(f.event)}</strong><div class="spacer"></div>${badge(f.decision || "event")}</div>
+          <div class="t-meta">${escapeHtml(String(f.actor||""))} · ${escapeHtml(f.when||"")}</div>
+          <pre>${escapeHtml(JSON.stringify({ why: f.why, input: f.input, output: f.output }, null, 2))}</pre>
+        </div>`).join("") || `<div class="empty">No replay frames.</div>`;
+      } catch (err) { out.innerHTML = `<div class="empty">${escapeHtml(err.message)}</div>`; }
+    });
+
     loadRoles().catch((e) => toast(e.message));
     loadRuns().catch(() => {});
   </script>

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/plugins.html

@@ -12,6 +12,16 @@
     <p class="sub" id="sub">Versioned, permissioned plugins that extend skills, tools, and workflows.</p>
     <div id="list" class="grid"><div class="empty">Loading plugins…</div></div>
 
+    <div class="section">
+      <h3>Template foundation</h3>
+      <div id="templates" class="grid"><div class="empty">Loading templates…</div></div>
+    </div>
+
+    <div class="section">
+      <h3>Plugin execution viewer</h3>
+      <div id="pluginEvents"><div class="empty">Loading plugin events…</div></div>
+    </div>
+
     <div class="section">
       <h3>Validate a manifest</h3>
       <p class="sub">Paste a <code>plugin.json</code> to check it against the SDK schema and permission allow-list.</p>
@@ -55,6 +65,45 @@
         </div>`).join("");
     }
 
+    async function loadTemplates() {
+      const data = await api("/marketplace/templates");
+      const box = document.getElementById("templates");
+      if (!data.templates.length) { box.innerHTML = `<div class="empty">No templates available.</div>`; return; }
+      box.innerHTML = data.templates.map((t) => `
+        <div class="card">
+          <div class="row"><h3>${escapeHtml(t.name)}</h3><div class="spacer"></div>${badge(t.kind)}</div>
+          <div class="meta">v${escapeHtml(t.version)} · ${escapeHtml((t.metadata||{}).category || "foundation")}</div>
+          <p style="font-size:13px;color:#cbd5e1">${escapeHtml(t.description || "")}</p>
+          <div class="row" style="margin-top:12px">
+            <a class="btn ghost" target="_blank" href="/marketplace/templates/${t.kind}/${t.id}/export">Export</a>
+            <button data-template="${t.kind}:${t.id}">Install</button>
+          </div>
+        </div>`).join("");
+    }
+
+    async function loadPluginEvents() {
+      const data = await api("/realtime/feed?limit=80");
+      const events = (data.events || []).filter((e) => e.area === "plugins" || e.event_type?.startsWith("plugin_"));
+      const box = document.getElementById("pluginEvents");
+      if (!events.length) { box.innerHTML = `<div class="empty">No plugin executions yet.</div>`; return; }
+      box.innerHTML = events.slice(0, 20).map((ev) => `<div class="timeline-item">
+        <div class="row"><strong>${escapeHtml(ev.event_type)}</strong><div class="spacer"></div>${badge((ev.payload||{}).status || ev.area)}</div>
+        <div class="t-meta">${escapeHtml((ev.payload||{}).plugin_id || (ev.payload||{}).plugin || "")} · ${escapeHtml(ev.received_at || ev.timestamp || "")}</div>
+      </div>`).join("");
+    }
+
+    document.getElementById("templates").addEventListener("click", async (e) => {
+      const btn = e.target.closest("button[data-template]");
+      if (!btn) return;
+      btn.disabled = true;
+      const [kind, id] = btn.dataset.template.split(":");
+      try {
+        const exported = await api(`/marketplace/templates/${kind}/${id}/export`);
+        await api("/marketplace/templates/install", { method: "POST", body: JSON.stringify({ data: exported }) });
+        toast(`Installed template: ${id}`);
+      } catch (err) { toast(err.message); } finally { btn.disabled = false; }
+    });
+
     document.getElementById("list").addEventListener("click", async (e) => {
       const btn = e.target.closest("button[data-act]");
       if (!btn) return;
@@ -77,6 +126,8 @@
     });
 
     load().catch((e) => toast(e.message));
+    loadTemplates().catch((e) => toast(e.message));
+    loadPluginEvents().catch(() => {});
   </script>
 </body>
 </html>