fix: UI/UX 修复 + 暗色主题 + async generator 防御

- App.vue: 重构 bootstrapBackend 流程，新增 retryBootstrap 重试入口
- SplashScreen.vue: 错误状态显示「重试」按钮
- system.py: /system/resources 移除 SYSTEM_CONFIG 权限依赖，避免 dev 模式 401
- react.py + gateway.py: 新增 _ensure_async_iterable helper 防御
  'async for requires aiter, got coroutine'
- theme.ts: Ant Design colorTextLightSolid 映射到 --text-inverse
  修复暗色主题下所有 primary 按钮白底白字
- ChatSidebar.vue: 新建对话按钮兜底深色文字
- SystemMonitorPanel.vue: 服务状态区域间距优化
- chat.ts + portal.py + sqlite_conversation_store.py: 会话标题派生修复
  解决点击对话标题变成"对话"的问题
- app.py: Serve 模式自动创建 default agent
- Tauri src-tauri/: 完整 Tauri 客户端配置 (icons, capabilities, Cargo)

src/agentkit/chat/sqlite_conversation_store.py

@@ -303,3 +303,36 @@ class SqliteConversationStore:
     ) -> None:
         """No-op for SQLite store — data is already persisted in the database."""
         # Nothing to do; all data lives in SQLite and is loaded on demand.
+
+    async def get_first_user_message(self, conversation_id: str) -> ChatMessage | None:
+        """Get the first user message of a conversation (authoritative source for title).
+
+        Reads directly from SQLite so the title is correct even when the
+        in-memory cache has not been populated or has an empty `messages`
+        list (e.g. after a server restart).
+        """
+        db = await self._ensure_db()
+        try:
+            cursor = await db.execute(
+                "SELECT role, content, timestamp, metadata FROM messages "
+                "WHERE conversation_id = ? AND role = 'user' AND content != '' "
+                "ORDER BY id ASC LIMIT 1",
+                (conversation_id,),
+            )
+            row = await cursor.fetchone()
+            if not row:
+                return None
+            meta: dict[str, Any] = {}
+            try:
+                meta = json.loads(row["metadata"]) if row["metadata"] else {}
+            except (json.JSONDecodeError, TypeError):
+                pass
+            return ChatMessage(
+                role=row["role"],
+                content=row["content"],
+                timestamp=self._str_to_dt(row["timestamp"]),
+                metadata=meta,
+            )
+        except Exception as e:
+            logger.warning(f"get_first_user_message failed for {conversation_id}: {e}")
+            return None

src/agentkit/core/react.py

@@ -45,6 +45,71 @@ class ReActStep:
     tokens: int = 0
 
 
+async def _ensure_async_iterable(obj: Any, label: str = "<obj>"):
+    """Defensive helper: ensure the given object is an async iterable.
+
+    Guards against the recurring ``'async for' requires an object with
+    __aiter__ method, got coroutine`` error. This error happens when an
+    ``async def`` function that *should* yield values ends up returning a
+    coroutine object instead of an async generator — typically because
+    every code path through the function exits before the first ``yield``
+    (e.g. early ``raise``) and a misbehaving caller in some Python
+    versions or a specific runtime configuration treats it as a coroutine.
+
+    This helper accepts either:
+    - An async iterable (async generator) → returned as-is.
+    - An awaitable that resolves to an async iterable → awaited, then yielded.
+    - Anything else → raises a clear, actionable error naming ``label``.
+
+    Use it like::
+
+        async for chunk in _ensure_async_iterable(
+            some_func_that_should_stream(), label="some_func"
+        ):
+            ...
+
+    Args:
+        obj: The object returned by calling an ``async def`` function.
+        label: A short human-readable name used in error messages to help
+            locate the source of the bug.
+
+    Yields:
+        Items from the resolved async iterable.
+
+    Raises:
+        TypeError: If ``obj`` is neither an async iterable nor an
+            awaitable that resolves to one. The error message names
+            ``label`` so the offending call site is easy to find.
+    """
+    # Case 1: already an async iterable (the normal case).
+    if hasattr(obj, "__aiter__"):
+        async for item in obj:
+            yield item
+        return
+
+    # Case 2: an awaitable that hasn't been awaited yet (the bug we're
+    # guarding against). Awaiting it should produce an async iterable.
+    if asyncio.iscoroutine(obj) or asyncio.isfuture(obj):
+        resolved = await obj
+        if hasattr(resolved, "__aiter__"):
+            async for item in resolved:
+                yield item
+            return
+        raise TypeError(
+            f"{label}: awaited value is not async iterable "
+            f"(got {type(resolved).__name__})"
+        )
+
+    # Case 3: anything else — surface a clear, actionable error rather
+    # than the cryptic CPython ``TypeError: 'async for' requires...``.
+    raise TypeError(
+        f"{label}: expected an async iterable, got {type(obj).__name__}. "
+        f"This usually means the called function returned a coroutine "
+        f"instead of an async generator — check that it contains at "
+        f"least one reachable ``yield`` statement."
+    )
+
+
 @dataclass
 class ReActResult:
     """ReAct 执行结果"""
@@ -911,12 +976,15 @@ class ReActEngine:
                 stream_tool_calls: list[Any] = []
                 stream_model = model
 
-                async for chunk in self._llm_gateway.chat_stream(
+                async for chunk in _ensure_async_iterable(
+                    self._llm_gateway.chat_stream(
                         messages=conversation,
                         model=model,
                         agent_name=agent_name,
                         task_type=task_type,
                         tools=tool_schemas,
+                    ),
+                    label=f"llm_gateway.chat_stream(model={model!r})",
                 ):
                     if chunk.content:
                         stream_content_chunks.append(chunk.content)

src/agentkit/llm/gateway.py

@@ -1,5 +1,6 @@
 """LLM Gateway - 统一 LLM 调用入口"""
 
+import asyncio
 import logging
 import time
 from typing import Any
@@ -317,7 +318,28 @@ class LLMGateway:
             final_model = model_name
 
             try:
-                async for chunk in provider.chat_stream(stream_request):
+                stream_obj = provider.chat_stream(stream_request)
+                # Defensive: guard against misconfigured providers (e.g. an
+                # AsyncMock in tests, or a future refactor that accidentally
+                # turns chat_stream into a regular ``async def``) that return
+                # a coroutine instead of an async generator. The original
+                # cryptic error ``'async for' requires an object with
+                # __aiter__ method, got coroutine`` becomes a clear,
+                # actionable message naming the offending provider+model.
+                if asyncio.iscoroutine(stream_obj):
+                    logger.error(
+                        f"Provider '{model_name}'.chat_stream returned a "
+                        f"coroutine instead of an async generator. "
+                        f"Check that the method is defined as "
+                        f"``async def chat_stream(...): ...; yield ...``."
+                    )
+                    raise TypeError(
+                        f"Provider '{model_name}' returned a coroutine "
+                        f"from chat_stream() — expected an async "
+                        f"generator. This indicates a provider "
+                        f"implementation bug."
+                    )
+                async for chunk in stream_obj:
                     chunk_yielded = True
                     if chunk.content:
                         total_content += chunk.content

src/agentkit/server/app.py

@@ -168,9 +168,10 @@ async def lifespan(app: FastAPI):
 
     await _conversation_store.restore_from_store()
 
-    # In GUI mode, ensure a default chat agent exists with memory + tools
-    gui_mode = os.environ.get("AGENTKIT_GUI_MODE")
-    if gui_mode and not app.state.agent_pool.list_agents():
+    # Ensure a default chat agent exists with memory + tools (both GUI and serve modes)
+    # Previously this only ran in GUI mode, which caused 'Agent default not found' errors
+    # when the frontend tried to create a chat session via REST/WebSocket on a serve-mode server.
+    if not app.state.agent_pool.list_agents():
         from agentkit.core.config_driven import AgentConfig
         from agentkit.memory.profile import MemoryStore
         from agentkit.tools.memory_tool import MemoryTool

src/agentkit/server/frontend/src-tauri/.gitignore

@@ -0,0 +1,4 @@
+# Generated by Cargo
+# will have compiled files and executables
+/target/
+/gen/schemas

src/agentkit/server/frontend/src-tauri/Cargo.toml

@@ -0,0 +1,26 @@
+[package]
+name = "app"
+version = "0.1.0"
+description = "A Tauri App"
+authors = ["you"]
+license = ""
+repository = ""
+edition = "2021"
+rust-version = "1.77.2"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[lib]
+name = "app_lib"
+crate-type = ["staticlib", "cdylib", "rlib"]
+
+[build-dependencies]
+tauri-build = { version = "2.6.2", features = [] }
+
+[dependencies]
+serde_json = "1.0"
+serde = { version = "1.0", features = ["derive"] }
+log = "0.4"
+ureq = { version = "2.10", features = ["json"] }
+tauri = { version = "2.11.2", features = [] }
+tauri-plugin-log = "2"

src/agentkit/server/frontend/src-tauri/build.rs

@@ -0,0 +1,3 @@
+fn main() {
+  tauri_build::build()
+}

src/agentkit/server/frontend/src-tauri/capabilities/default.json

@@ -0,0 +1,11 @@
+{
+  "$schema": "../gen/schemas/desktop-schema.json",
+  "identifier": "default",
+  "description": "enables the default permissions",
+  "windows": [
+    "main"
+  ],
+  "permissions": [
+    "core:default"
+  ]
+}

src/agentkit/server/frontend/src-tauri/src/lib.rs

@@ -0,0 +1,131 @@
+use std::sync::Mutex;
+
+/// Tauri 2.x IPC commands for AgentKit desktop client.
+///
+/// Architecture: Tauri client connects to a remote AgentKit server (not a
+/// local sidecar). The Rust shell acts as a thin shim that exposes the
+/// remote server's connection info to the frontend, and proxies health
+/// checks through the local network.
+///
+/// Frontend contract (see src/api/tauri.ts):
+///   - start_backend()           -> u16  (remote port the frontend should use)
+///   - get_backend_port()        -> u16
+///   - stop_backend()            -> ()
+///   - check_backend_health()    -> bool (proxied HTTP GET to remote /api/v1/health)
+///   - get_backend_url()         -> String (full base URL, e.g. http://127.0.0.1:8000)
+
+/// Remote server connection info. Defaults match the local dev server
+/// (``agentkit serve --port 8000``). Override at compile time via
+/// ``AGENTKIT_REMOTE_HOST`` / ``AGENTKIT_REMOTE_PORT``.
+const REMOTE_HOST: &str = match option_env!("AGENTKIT_REMOTE_HOST") {
+    Some(h) => h,
+    None => "127.0.0.1",
+};
+const REMOTE_PORT_STR: &str = match option_env!("AGENTKIT_REMOTE_PORT") {
+    Some(p) => p,
+    None => "8000",
+};
+
+/// Parse the port at first use (avoids non-const fn call in constant).
+fn remote_port() -> u16 {
+    REMOTE_PORT_STR.parse().unwrap_or(8000)
+}
+
+/// Global state: tracks whether the "backend" is considered started.
+/// In our remote-server model this is a one-shot init flag.
+#[derive(Default)]
+struct BackendState {
+    started: Mutex<bool>,
+}
+
+/// Start the backend connection. Returns the remote server's port.
+///
+/// In a local-sidecar design this would spawn the Python process and
+/// return its bound port. In the remote-server model we simply record
+/// that initialization completed and return the configured remote port.
+#[tauri::command]
+fn start_backend(state: tauri::State<BackendState>) -> Result<u16, String> {
+    let mut started = state.started.lock().map_err(|e| e.to_string())?;
+    *started = true;
+    let port = remote_port();
+    log::info!("start_backend: connecting to remote server {REMOTE_HOST}:{port}");
+    Ok(port)
+}
+
+/// Get the current backend port. Errors if start_backend hasn't been
+/// called yet (matches the original sidecar contract).
+#[tauri::command]
+fn get_backend_port(state: tauri::State<BackendState>) -> Result<u16, String> {
+    let started = state.started.lock().map_err(|e| e.to_string())?;
+    if !*started {
+        return Err("Backend not started. Call start_backend() first.".to_string());
+    }
+    Ok(remote_port())
+}
+
+/// Stop the backend connection. In the remote-server model this is a
+/// no-op (we don't own the server lifecycle) but we still clear the
+/// started flag so a subsequent start_backend() can re-initialize.
+#[tauri::command]
+fn stop_backend(state: tauri::State<BackendState>) -> Result<(), String> {
+    let mut started = state.started.lock().map_err(|e| e.to_string())?;
+    *started = false;
+    log::info!("stop_backend: cleared connection state");
+    Ok(())
+}
+
+/// Probe the remote server's health endpoint. Returns true if it
+/// responds with 200 OK to ``GET /api/v1/health`` within 3 seconds.
+///
+/// We use a synchronous HTTP client (ureq) here to keep the Tauri
+/// command body simple and avoid pulling in tokio/reqwest for a single
+/// health probe. The probe runs in a worker thread automatically.
+#[tauri::command]
+fn check_backend_health() -> Result<bool, String> {
+    let url = format!("http://{REMOTE_HOST}:{}/api/v1/health", remote_port());
+    log::debug!("check_backend_health: GET {url}");
+    match ureq::get(&url).timeout(std::time::Duration::from_secs(3)).call() {
+        Ok(resp) if resp.status() == 200 => Ok(true),
+        Ok(resp) => {
+            log::warn!("check_backend_health: non-200 status {}", resp.status());
+            Ok(false)
+        }
+        Err(e) => {
+            log::warn!("check_backend_health: {e}");
+            Ok(false)
+        }
+    }
+}
+
+/// Return the full base URL the frontend should use to reach the
+/// remote server. Convenience command that the frontend could use
+/// instead of constructing the URL from get_backend_port().
+#[tauri::command]
+fn get_backend_url() -> Result<String, String> {
+    Ok(format!("http://{REMOTE_HOST}:{}", remote_port()))
+}
+
+#[cfg_attr(mobile, tauri::mobile_entry_point)]
+pub fn run() {
+    tauri::Builder::default()
+        .manage(BackendState::default())
+        .setup(|app| {
+            if cfg!(debug_assertions) {
+                app.handle().plugin(
+                    tauri_plugin_log::Builder::default()
+                        .level(log::LevelFilter::Info)
+                        .build(),
+                )?;
+            }
+            Ok(())
+        })
+        .invoke_handler(tauri::generate_handler![
+            start_backend,
+            get_backend_port,
+            stop_backend,
+            check_backend_health,
+            get_backend_url,
+        ])
+        .run(tauri::generate_context!())
+        .expect("error while running tauri application");
+}

src/agentkit/server/frontend/src-tauri/src/main.rs

@@ -0,0 +1,6 @@
+// Prevents additional console window on Windows in release, DO NOT REMOVE!!
+#![cfg_attr(not(debug_assertions), windows_subsystem = "windows")]
+
+fn main() {
+  app_lib::run();
+}

src/agentkit/server/frontend/src-tauri/tauri.conf.json

@@ -0,0 +1,40 @@
+{
+  "$schema": "../node_modules/@tauri-apps/cli/config.schema.json",
+  "productName": "AgentKit",
+  "version": "0.1.0",
+  "identifier": "com.fischer.agentkit",
+  "build": {
+    "frontendDist": "../static",
+    "devUrl": "http://localhost:5173",
+    "beforeDevCommand": "npm run dev",
+    "beforeBuildCommand": "npm run build:frontend"
+  },
+  "app": {
+    "windows": [
+      {
+        "title": "AgentKit Desktop",
+        "width": 1280,
+        "height": 800,
+        "minWidth": 1024,
+        "minHeight": 600,
+        "resizable": true,
+        "fullscreen": false,
+        "center": true
+      }
+    ],
+    "security": {
+      "csp": null
+    }
+  },
+  "bundle": {
+    "active": true,
+    "targets": "all",
+    "icon": [
+      "icons/32x32.png",
+      "icons/128x128.png",
+      "icons/128x128@2x.png",
+      "icons/icon.icns",
+      "icons/icon.ico"
+    ]
+  }
+}

src/agentkit/server/frontend/src/App.vue

@@ -1,6 +1,11 @@
 <template>
   <a-config-provider :locale="zhCN" :theme="themeStore.antThemeConfig">
-    <SplashScreen v-if="loading" :status="loadingStatus" :error="loadError" />
+    <SplashScreen
+      v-if="loading"
+      :status="loadingStatus"
+      :error="loadError"
+      :on-retry="retryBootstrap"
+    />
     <router-view v-else />
   </a-config-provider>
 </template>
@@ -41,17 +46,7 @@ onMounted(async () => {
 
   if (isTauri()) {
     try {
-      loadingStatus.value = '正在启动后端...'
-      await startBackend()
-      loadingStatus.value = '正在配置连接...'
-      await initApiBaseURL()
-      loadingStatus.value = '正在检查服务...'
-      const healthy = await checkBackendHealth()
-      if (healthy) {
-        loading.value = false
-      } else {
-        loadError.value = '后端服务健康检查失败'
-      }
+      await bootstrapBackend()
     } catch (err: unknown) {
       loadError.value = err instanceof Error ? err.message : String(err)
     }
@@ -59,6 +54,37 @@ onMounted(async () => {
     loading.value = false
   }
 })
+
+/** Run the Tauri-side backend bootstrap (start + init URL + health check).
+ *
+ * Exposed to the template so the user can click "重试" when the initial
+ * health check fails (e.g. the Python server was not up at the moment
+ * the window opened). */
+async function bootstrapBackend(): Promise<void> {
+  loadingStatus.value = '正在启动后端...'
+  await startBackend()
+  loadingStatus.value = '正在配置连接...'
+  await initApiBaseURL()
+  loadingStatus.value = '正在检查服务...'
+  const healthy = await checkBackendHealth()
+  if (healthy) {
+    loadError.value = ''
+    loading.value = false
+  } else {
+    loadError.value = '后端服务健康检查失败，请确认 agentkit serve 已在 8000 端口运行后点击重试。'
+  }
+}
+
+/** User-triggered retry from the error screen. */
+async function retryBootstrap(): Promise<void> {
+  loadError.value = ''
+  loading.value = true
+  try {
+    await bootstrapBackend()
+  } catch (err: unknown) {
+    loadError.value = err instanceof Error ? err.message : String(err)
+  }
+}
 </script>
 
 <style>

src/agentkit/server/frontend/src/components/chat/ChatSidebar.vue

@@ -105,6 +105,22 @@ function formatRelativeTime(dateStr: string): string {
   border-bottom: 1px solid var(--border-color-split);
 }
 
+/* Fix: dark theme — primary button has light bg (--color-primary=#fbfbfa)
+ * in dark mode, so default white text is invisible. Force dark text in
+ * dark theme to keep "新建对话" readable. */
+.chat-sidebar__header :deep(.ant-btn-primary) {
+  color: var(--text-inverse);
+}
+
+[data-theme="dark"] .chat-sidebar__header :deep(.ant-btn-primary) {
+  color: #1a1a1a;
+}
+
+[data-theme="dark"] .chat-sidebar__header :deep(.ant-btn-primary:hover),
+[data-theme="dark"] .chat-sidebar__header :deep(.ant-btn-primary:focus) {
+  color: #1a1a1a;
+}
+
 .chat-sidebar__list {
   flex: 1;
   overflow-y: auto;

src/agentkit/server/frontend/src/components/layout/SplashScreen.vue

@@ -3,13 +3,16 @@
     <div class="splash-content">
       <div class="splash-logo">Fischer AgentKit</div>
       <div class="splash-subtitle">AI Agent Framework</div>
-      <div class="splash-progress">
+      <div v-if="!error" class="splash-progress">
         <div class="splash-progress-bar">
           <div class="splash-progress-inner"></div>
         </div>
       </div>
-      <div class="splash-status">{{ status }}</div>
-      <div v-if="error" class="splash-error">{{ error }}</div>
+      <div v-if="!error" class="splash-status">{{ status }}</div>
+      <div v-if="error" class="splash-error">
+        <div class="splash-error-text">{{ error }}</div>
+        <button v-if="onRetry" class="splash-retry" @click="onRetry">重试</button>
+      </div>
     </div>
   </div>
 </template>
@@ -18,6 +21,7 @@
 defineProps<{
   status: string
   error?: string
+  onRetry?: () => void | Promise<void>
 }>()
 </script>
 
@@ -74,9 +78,33 @@ defineProps<{
 }
 .splash-error {
   margin-top: 12px;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 12px;
+}
+.splash-error-text {
   font-size: 12px;
   color: var(--color-error, #ff4d4f);
-  max-width: 300px;
+  max-width: 320px;
   text-align: center;
+  line-height: 1.5;
+}
+.splash-retry {
+  padding: 6px 18px;
+  font-size: 13px;
+  font-weight: 500;
+  color: #fff;
+  background: var(--color-primary, #6366f1);
+  border: 1px solid var(--color-primary, #6366f1);
+  border-radius: 6px;
+  cursor: pointer;
+  transition: opacity 0.15s ease, transform 0.1s ease;
+}
+.splash-retry:hover {
+  opacity: 0.85;
+}
+.splash-retry:active {
+  transform: scale(0.97);
 }
 </style>

src/agentkit/server/frontend/src/components/layout/SystemMonitorPanel.vue

@@ -232,10 +232,11 @@ onUnmounted(() => {
 .system-monitor__tabs {
   display: flex;
   gap: var(--space-4);
-  padding: 12px 16px;
+  padding: 14px 16px;
   border-bottom: 1px solid var(--border-color);
   font-size: var(--font-sm);
   color: var(--text-tertiary);
+  flex-shrink: 0;
 }
 
 .system-monitor__tab {
@@ -243,7 +244,7 @@ onUnmounted(() => {
   align-items: center;
   gap: 5px;
   cursor: pointer;
-  padding: 4px 0;
+  padding: 6px 0;
   border: none;
   border-bottom: 2px solid transparent;
   background: transparent;
@@ -286,43 +287,69 @@ onUnmounted(() => {
 .system-monitor__body {
   flex: 1;
   overflow-y: auto;
-  padding: 16px;
+  padding: 18px 16px;
 }
 
 .system-monitor__section-title {
   font-size: var(--font-sm);
   font-weight: var(--font-weight-semibold);
   color: var(--text-primary);
-  margin: 16px 0 10px;
+  margin: 18px 0 12px;
+  display: flex;
+  align-items: center;
+  gap: 6px;
 }
 
 .system-monitor__section-title:first-child {
   margin-top: 0;
 }
 
+.system-monitor__section-title::before {
+  content: '';
+  display: inline-block;
+  width: 3px;
+  height: 12px;
+  background: var(--accent-team);
+  border-radius: 2px;
+}
+
+/* 系统概览区：4 个指标卡片 + 分隔线 */
 .system-monitor__metrics {
   display: grid;
   grid-template-columns: 1fr 1fr;
-  gap: 8px;
+  gap: 10px;
+  padding-bottom: 22px;
+  margin-bottom: 4px;
+  border-bottom: 1px solid var(--border-color);
 }
 
 .system-monitor__metric {
-  padding: 12px;
+  padding: 14px 12px;
   border: 1px solid var(--border-color);
   border-radius: var(--radius-md);
+  background: var(--bg-secondary);
+  transition: border-color 0.15s;
+}
+
+.system-monitor__metric:hover {
+  border-color: var(--color-primary-light);
 }
 
 .system-monitor__metric-label {
   font-size: 11px;
   color: var(--text-tertiary);
-  margin-bottom: 4px;
+  margin-bottom: 6px;
+  display: flex;
+  align-items: center;
+  gap: 4px;
 }
 
 .system-monitor__metric-value {
-  font-size: 18px;
+  font-size: 20px;
   font-weight: var(--font-weight-semibold);
   color: var(--text-primary);
-  margin-bottom: 2px;
+  margin-bottom: 4px;
+  line-height: 1.2;
 }
 
 .system-monitor__metric-value.status-healthy {
@@ -341,6 +368,7 @@ onUnmounted(() => {
 .system-monitor__metric-delta {
   font-size: 10px;
   color: var(--text-tertiary);
+  line-height: 1.4;
 }
 
 .system-monitor__metric-delta.up {
@@ -351,28 +379,46 @@ onUnmounted(() => {
   color: var(--accent-team);
 }
 
+/* 分区容器：左右两列独立分块 */
 .system-monitor__split {
   display: grid;
   grid-template-columns: 1fr;
-  gap: 16px;
+  gap: 22px;
+  margin-top: 6px;
 }
 
+.system-monitor__column {
+  display: flex;
+  flex-direction: column;
+  min-width: 0;
+}
+
+/* 服务状态区：与上方 metrics 区域有明显的呼吸间距 */
 .system-monitor__services {
   display: flex;
   flex-direction: column;
-  gap: 2px;
+  gap: 0;
+  padding: 6px 14px;
+  background: var(--bg-secondary);
+  border-radius: var(--radius-md);
+  border: 1px solid var(--border-color);
+  margin-top: 4px;
 }
 
 .system-monitor__service {
   display: grid;
   grid-template-columns: 10px minmax(0, 1fr) auto auto;
   align-items: center;
-  gap: 8px;
-  padding: 5px 0;
+  gap: 10px;
+  padding: 10px 0;
   font-size: var(--font-sm);
   border-bottom: 1px solid var(--border-color);
 }
 
+.system-monitor__service:last-child {
+  border-bottom: none;
+}
+
 .system-monitor__service-name,
 .system-monitor__service-status,
 .system-monitor__service-time {
@@ -382,27 +428,34 @@ onUnmounted(() => {
 }
 
 .system-monitor__service-dot {
-  width: 6px;
-  height: 6px;
+  width: 8px;
+  height: 8px;
   border-radius: 50%;
   background: var(--text-placeholder);
+  box-shadow: 0 0 0 2px var(--bg-secondary);
 }
 
 .system-monitor__service-dot.ok {
   background: var(--color-success);
+  box-shadow: 0 0 0 2px rgba(82, 196, 26, 0.12);
 }
 
 .system-monitor__service-dot.error {
   background: var(--color-error);
+  box-shadow: 0 0 0 2px rgba(245, 34, 45, 0.12);
 }
 
 .system-monitor__service-name {
   color: var(--text-primary);
+  font-weight: var(--font-weight-medium, 500);
 }
 
 .system-monitor__service-status {
   color: var(--color-success);
   font-size: 11px;
+  padding: 1px 6px;
+  border-radius: 3px;
+  background: rgba(82, 196, 26, 0.08);
 }
 
 .system-monitor__service-time {
@@ -412,7 +465,7 @@ onUnmounted(() => {
 }
 
 .system-monitor__chart-placeholder {
-  height: 80px;
+  min-height: 96px;
   display: flex;
   flex-direction: column;
   align-items: center;
@@ -422,6 +475,8 @@ onUnmounted(() => {
   font-size: var(--font-xs);
   background: var(--bg-secondary);
   border-radius: var(--radius-md);
-  margin: 8px 0;
+  border: 1px dashed var(--border-color);
+  margin-top: 4px;
+  padding: 12px;
 }
 </style>

src/agentkit/server/frontend/src/stores/chat.ts

@@ -75,16 +75,30 @@ export const useChatStore = defineStore('chat', () => {
         const fullConv = await apiClient.getConversation(id)
         if (conv) {
           conv.messages = fullConv.messages || []
-          conv.title = fullConv.title || conv.title
+          // P0 #1 fix: never let the server's placeholder title ("对话")
+          // overwrite a real title we already have locally. If the server
+          // returns the placeholder, keep whatever the list view had.
+          const serverTitle = fullConv.title || ''
+          const localTitle = conv.title || ''
+          const isServerPlaceholder =
+            serverTitle === '对话' || serverTitle.trim() === ''
+          const isLocalReal = localTitle && localTitle !== '新对话' && localTitle !== '对话'
+          if (serverTitle && !isServerPlaceholder) {
+            conv.title = serverTitle
+          } else if (!isLocalReal) {
+            // Both sides have no real title — accept server's placeholder
+            conv.title = serverTitle || localTitle || '对话'
+          }
           conv.created_at = fullConv.created_at || conv.created_at
           conv.updated_at = fullConv.updated_at || conv.updated_at
         } else {
           // P1 #7 fix: If the conversation is not in the local list (e.g.
           // after a page refresh), add it from the fetched data instead
           // of silently discarding the result.
+          const serverTitle = fullConv.title || '新对话'
           conversations.value.unshift({
             id: fullConv.id || id,
-            title: fullConv.title || '新对话',
+            title: serverTitle === '对话' ? '新对话' : serverTitle,
             messages: fullConv.messages || [],
             created_at: fullConv.created_at || new Date().toISOString(),
             updated_at: fullConv.updated_at || new Date().toISOString(),

src/agentkit/server/frontend/src/stores/theme.ts

@@ -100,6 +100,14 @@ export const useThemeStore = defineStore('theme', () => {
         colorTextSecondary: readToken('--text-secondary', isDark ? '#cececd' : '#4a4a4a'),
         colorTextTertiary: readToken('--text-tertiary', isDark ? '#9b9b9a' : '#6b6b6a'),
         colorTextQuaternary: readToken('--text-placeholder', isDark ? '#6b6b6a' : '#9b9b9a'),
+        // Text rendered ON top of ``colorPrimary`` (primary buttons, active
+        // tab indicators, etc.). Ant Design hardcodes this to #fff in both
+        // light and dark themes, which clashes with our design: in dark
+        // mode ``--color-primary`` is a near-white pill (#fbfbfa) and we
+        // want dark text on it. Source the value from --text-inverse so
+        // light and dark themes stay consistent with the rest of the
+        // design tokens.
+        colorTextLightSolid: readToken('--text-inverse', isDark ? '#1a1a1a' : '#ffffff'),
         colorBgContainer: readToken('--bg-primary', isDark ? '#1a1a1a' : '#ffffff'),
         colorBgLayout: readToken('--bg-secondary', isDark ? '#1f1f1f' : '#fbfbfa'),
         colorBgElevated: readToken('--bg-elevated', isDark ? '#252525' : '#ffffff'),

src/agentkit/server/routes/portal.py

@@ -656,40 +656,71 @@ async def get_capabilities(req: Request, _auth: None = Depends(_verify_api_key))
 
 @router.get("/portal/conversations")
 async def list_conversations(limit: int = 20, _auth: None = Depends(_verify_api_key)):
-    """List recent conversations."""
+    """List recent conversations.
+
+    For each conversation, derive the title from the first user message
+    read directly from SQLite (independent of the in-memory cache, which
+    may have an empty `messages` list after a restart). This prevents the
+    regression where titles collapse to the placeholder "对话".
+    """
     convs = await _conversation_store.list_conversations(limit=limit)
-    return [
+    result: list[dict] = []
+    for c in convs:
+        # Re-derive title from the persisted user message so cache misses
+        # after a restart don't surface the default placeholder.
+        first_user = await _conversation_store.get_first_user_message(c.id)
+        title = _derive_conversation_title_from_content(
+            first_user.content if first_user else None
+        )
+        result.append(
             {
                 "id": c.id,
-            "title": _derive_conversation_title(c),
+                "title": title,
                 "created_at": c.created_at.isoformat(),
                 "updated_at": c.updated_at.isoformat(),
                 "message_count": len(c.messages),
             }
-        for c in convs
-    ]
+        )
+    return result
 
 
 def _derive_conversation_title(conv: Conversation) -> str:
-    """Derive a human-readable title from the first user message."""
+    """Derive a human-readable title from the first user message in the conversation object."""
     for msg in conv.messages:
         if msg.role == "user" and msg.content:
             return msg.content[:20] + ("..." if len(msg.content) > 20 else "")
     return "对话"
 
 
+def _derive_conversation_title_from_content(content: str | None) -> str:
+    """Derive title from a string content (used when conv.messages is empty)."""
+    if content:
+        return content[:20] + ("..." if len(content) > 20 else "")
+    return "对话"
+
+
 @router.get("/portal/conversations/{conversation_id}")
 async def get_conversation(
     conversation_id: str, limit: int = 50, _auth: None = Depends(_verify_api_key)
 ):
-    """Get conversation history from SQLite-backed store."""
+    """Get conversation history from SQLite-backed store.
+
+    Title is derived from the first user message in `history` (the
+    authoritative source) rather than the in-memory cache, which may have
+    an empty `messages` list after a server restart. This prevents the
+    regression where selecting a conversation collapses the title to "对话".
+    """
     history = await _conversation_store.get_history(conversation_id, limit=limit)
     if not history:
         raise HTTPException(status_code=404, detail=f"Conversation '{conversation_id}' not found")
     conv = await _conversation_store.get_or_create(conversation_id)
+    first_user_content = next(
+        (m.content for m in history if m.role == "user" and m.content),
+        None,
+    )
     return {
         "id": conv.id,
-        "title": _derive_conversation_title(conv),
+        "title": _derive_conversation_title_from_content(first_user_content),
         "messages": [
             {
                 "id": f"{conv.id}-{i}",

src/agentkit/server/routes/system.py

@@ -6,10 +6,7 @@ import shutil
 import time
 from typing import Any
 
-from fastapi import APIRouter, Depends
-
-from agentkit.server.auth.dependencies import require_permission
-from agentkit.server.auth.permissions import Permission
+from fastapi import APIRouter
 
 logger = logging.getLogger(__name__)
 
@@ -32,15 +29,21 @@ def _read_meminfo() -> dict[str, int]:
     return values
 
 
-@router.get(
-    "/system/resources",
-    dependencies=[Depends(require_permission(Permission.SYSTEM_CONFIG))],
-)
+@router.get("/system/resources")
 async def get_system_resources() -> dict[str, Any]:
     """Return lightweight system resource usage.
 
     Uses only stdlib modules so it works without psutil. Values are best-effort
     and may be partial on platforms without /proc/meminfo.
+
+    Note: This endpoint is intentionally NOT gated by SYSTEM_CONFIG. It is
+    read-only monitoring data (CPU/memory/disk stats) that should be visible
+    to any authenticated or dev-mode user. Gating it under SYSTEM_CONFIG
+    caused 401 responses in dev mode, which the frontend's API client
+    treated as session expiration and triggered an unwanted redirect to
+    the login page. If stricter access control is required in the future,
+    add a dedicated ``SYSTEM_VIEW`` permission rather than reusing
+    ``SYSTEM_CONFIG``.
     """
     now = time.time()