feat: 完善知识图谱交互与 Agent/线程持久化

- 知识图谱画布、侧边栏、详情面板交互优化，新增图谱相关 API 与 hooks
- Agent 卡片与详情面板调整，对话页面重构
- 线程元数据持久化增强（base/memory/sql），knowledge 仓储与导出补充
- 网关路由 agents/knowledge/threads 同步更新

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

frontend-web/src/core/threads/hooks.ts

@@ -12,7 +12,7 @@ import { getAPIClient } from "../api";
 import { apiFetch } from "../api";
 import { getStoredAuth } from "../auth";
 import { rememberAskTime } from "../chat-timestamps";
-import { getBackendBaseURL } from "../config";
+import { getBackendBaseURL, getLangGraphBaseURL } from "../config";
 import { useI18n } from "../i18n/hooks";
 import type { FileInMessage } from "../messages/utils";
 import type { LocalSettings } from "../settings";
@@ -761,14 +761,86 @@ export function useThreads(
         offset += response.length;
       }
 
-      return threads.filter(
-        (t) => (t.metadata as Record<string, unknown> | null)?.thread_type !== "notebook",
-      );
+      // Safety net on top of the backend's system-thread exclusion: drop
+      // machine-created threads (scheduled tasks, roundtable workers,
+      // notebooks) in case an older backend still returns them.
+      return threads.filter((t) => {
+        const metadata = t.metadata as Record<string, unknown> | null;
+        if (metadata?.system) return false;
+        const threadType = metadata?.thread_type;
+        return (
+          threadType !== "notebook" &&
+          threadType !== "scheduler" &&
+          threadType !== "roundtable"
+        );
+      });
     },
     refetchOnWindowFocus: false,
   });
 }
 
+export interface ThreadsPageResult {
+  threads: AgentThread[];
+  hasMore: boolean;
+}
+
+/**
+ * Server-side paginated conversation list for the chats page. Unlike
+ * `useThreads` (which loads a bounded recent window for the sidebar), this
+ * fetches exactly one page per request so a long history is never pulled in
+ * one go. `search` matches thread titles on the backend; one extra row is
+ * over-fetched only to detect whether a next page exists. The query key stays
+ * under ["threads", "search"] so existing invalidations refresh pages too.
+ */
+export function useThreadsPage(options: {
+  page: number;
+  pageSize: number;
+  search?: string;
+  metadata?: Record<string, unknown>;
+}) {
+  const { page, pageSize, search, metadata } = options;
+  const normalizedSearch = search?.trim() ?? "";
+  return useQuery<ThreadsPageResult>({
+    queryKey: [
+      "threads",
+      "search",
+      "paged",
+      page,
+      pageSize,
+      normalizedSearch,
+      metadata ?? null,
+    ],
+    queryFn: async () => {
+      // Direct fetch instead of the LangGraph SDK: the SDK's threads.search
+      // drops fields it doesn't know about, and `query` (title search) is a
+      // gateway extension.
+      const response = await apiFetch(
+        `${getLangGraphBaseURL()}/threads/search`,
+        {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({
+            limit: pageSize + 1,
+            offset: (page - 1) * pageSize,
+            metadata: metadata ?? {},
+            query: normalizedSearch || undefined,
+          }),
+        },
+      );
+      if (!response.ok) {
+        throw new Error("Failed to load conversations");
+      }
+      const rows = (await response.json()) as AgentThread[];
+      return {
+        threads: rows.slice(0, pageSize),
+        hasMore: rows.length > pageSize,
+      };
+    },
+    placeholderData: (previous) => previous,
+    refetchOnWindowFocus: false,
+  });
+}
+
 export function useDeleteThread() {
   const queryClient = useQueryClient();
   const apiClient = getAPIClient();