Merge branch 'dhk-1' of http://192.168.0.200/knowledge-base-bj/deerflow-server into dhk-1

frontend-web/src/components/workspace/input-box.tsx

@@ -10,6 +10,7 @@ import {
   LightbulbIcon,
   PaperclipIcon,
   PlusIcon,
+  ShrinkIcon,
   SparklesIcon,
   RocketIcon,
   XIcon,
@@ -329,6 +330,14 @@ export function InputBox({
     setLocalSettings("display", { ragModeEnabled: !ragModeEnabled });
   }, [ragModeEnabled, ragModeLocked, setLocalSettings]);
 
+  const contextCompressionEnabled =
+    localSettings.display.contextCompressionEnabled ?? false;
+  const handleContextCompressionToggle = useCallback(() => {
+    setLocalSettings("display", {
+      contextCompressionEnabled: !contextCompressionEnabled,
+    });
+  }, [contextCompressionEnabled, setLocalSettings]);
+
   const handleSubmit = useCallback(
     async (message: PromptInputMessage) => {
       if (status === "streaming") {
@@ -735,6 +744,35 @@ export function InputBox({
                 <span className="text-xs font-normal">参考文献</span>
               </PromptInputButton>
             </Tooltip>
+            <Tooltip
+              content={
+                contextCompressionEnabled
+                  ? "上下文压缩已开启：超长对话会自动压缩历史以节省 token"
+                  : "上下文压缩已关闭：点击开启"
+              }
+            >
+              <PromptInputButton
+                type="button"
+                aria-pressed={contextCompressionEnabled}
+                disabled={disabled}
+                className={cn(
+                  "gap-1.5! rounded-full px-3! shadow-none",
+                  compact && "rounded-lg px-2.5! py-1! text-[11px] font-medium",
+                  contextCompressionEnabled
+                    ? "border border-violet-300 bg-violet-50 text-violet-700 hover:bg-violet-100 dark:border-violet-700 dark:bg-violet-950/40 dark:text-violet-200"
+                    : compact
+                      ? "border border-transparent bg-muted/40 text-muted-foreground hover:bg-muted hover:text-foreground"
+                      : "border-0 bg-transparent text-muted-foreground hover:bg-muted/70 hover:text-foreground dark:text-slate-400 dark:hover:bg-slate-700/70 dark:hover:text-slate-100",
+                )}
+                onClick={handleContextCompressionToggle}
+              >
+                {contextCompressionEnabled && (
+                  <CheckIcon className="size-3 text-violet-700 dark:text-violet-200" />
+                )}
+                <ShrinkIcon className="size-3" />
+                <span className="text-xs font-normal">上下文压缩</span>
+              </PromptInputButton>
+            </Tooltip>
             <PromptInputActionMenu>
               <ModeHoverGuide
                 mode={

frontend-web/src/core/knowledge/types.ts

@@ -147,6 +147,13 @@ export interface BatchIngestPayload {
   include_sources?: boolean;
   skip_existing?: boolean;
   mode?: "summary" | "full";
+  /**
+   * When true (default), the backend runs the per-thread extraction loop in a
+   * background task and returns immediately (`queued:true`) instead of blocking
+   * the request until every thread is processed — which otherwise outlives the
+   * nginx gateway timeout and surfaces as a generic "request failed".
+   */
+  background?: boolean;
   /** Optional model override for knowledge extraction (null = configured default). */
   model_name?: string | null;
   /** Optional extraction-prompt template id (null = scope default). */
@@ -157,6 +164,8 @@ export interface BatchIngestPayload {
 
 export interface BatchIngestResult {
   ok: boolean;
+  /** True when the backend queued the batch as a background task (see payload). */
+  queued?: boolean;
   processed: number;
   created: number;
   skipped: number;

frontend-web/src/core/page-layout/sidebar-menu.ts

@@ -208,12 +208,12 @@ export const sidebarMenuItems: MenuItem[] = [
         userPath: "/page/strategy/user-leaderboard",
       },
       { id: "mgmt-agents", label: "智能体管理", icon: Bot, path: "/page/strategy/agents" },
-      {
-        id: "mgmt-business-chain-mapping",
-        label: "配置管理",
-        icon: Network,
-        path: "/page/strategy/business-chain-mapping",
-      },
+      // {
+      //   id: "mgmt-business-chain-mapping",
+      //   label: "配置管理",
+      //   icon: Network,
+      //   path: "/page/strategy/business-chain-mapping",
+      // },
       {
         id: "mgmt-business-chains",
         label: "业务链条配置",

frontend-web/src/core/settings/local.ts

@@ -7,6 +7,7 @@ export const DEFAULT_LOCAL_SETTINGS: LocalSettings = {
   display: {
     streamSpeedMeterEnabled: true,
     ragModeEnabled: false,
+    contextCompressionEnabled: false,
     citationCleanupWords: [],
     fontScale: 1,
   },
@@ -43,6 +44,12 @@ export interface LocalSettings {
      * messages render tool results with the legacy generic search UI only.
      */
     ragModeEnabled: boolean;
+    /**
+     * Per-user switch for context compression (上下文压缩). When enabled, long
+     * conversations are summarized transiently before the model call to save
+     * tokens. Default off; toggled from the chat input box (below 参考文献).
+     */
+    contextCompressionEnabled: boolean;
     /**
      * Admin-maintained words or short phrases removed from reference snippets
      * and dialogs, one entry per line in the settings UI.

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

@@ -664,6 +664,9 @@ export function useThreadStream({
           lockThreadRagMode(threadId);
         }
 
+        const contextCompressionEnabled =
+          localSettings.display.contextCompressionEnabled ?? false;
+
         const mergedContext = {
           ...context,
           ...extraContext,
@@ -715,6 +718,7 @@ export function useThreadStream({
               memory_injection_enabled:
                 mergedContext.memory_injection_enabled ?? true,
               rag_mode_enabled: ragModeEnabled,
+              summarization_enabled: contextCompressionEnabled,
               thread_id: threadId,
             },
           },

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

@@ -20,6 +20,8 @@ export interface AgentThreadContext extends Record<string, unknown> {
   writing_artifact_path?: string;
   memory_injection_enabled?: boolean;
   rag_mode_enabled?: boolean;
+  /** Per-request switch for context compression (上下文压缩). Default off. */
+  summarization_enabled?: boolean;
   /** Per-conversation switch for knowledge-base RAG injection (phase 2). */
   knowledge_rag_enabled?: boolean;
   agent_id?: string;

frontend-web/src/pages/KnowledgeBasePage.tsx

@@ -176,10 +176,21 @@ function BatchImportDialog({
         skip_existing: skipExisting,
         status: "approved",
         include_sources: true,
+        // Run in the background so the request returns immediately — a large
+        // batch otherwise blocks past the gateway timeout and shows "request
+        // failed". Threads stream into the list as they finish.
+        background: true,
         model_name: model,
         template_id: template,
         folder,
       });
+      if (result.queued) {
+        toast.success(
+          `已转入后台导入 ${result.processed} 个线程，完成后会陆续出现在左侧列表（可稍后刷新查看）`,
+        );
+        onOpenChange(false);
+        return;
+      }
       const base = `批量导入完成：新增 ${result.created}，跳过 ${result.skipped}，失败 ${result.failed}`;
       if (result.llm_fallback && result.llm_fallback > 0) {
         // The model failed on some threads — they were saved via rule-based
@@ -549,7 +560,7 @@ export default function KnowledgeBasePage() {
       keyword: debouncedKeyword || undefined,
       source_type: sourceType === "all" ? undefined : sourceType,
       status: status === "all" ? undefined : status,
-      limit: 500,
+      limit: 1000,
     }),
     [debouncedKeyword, sourceType, status],
   );

offline-backend-20260512/backend/CLAUDE.md

@@ -252,7 +252,7 @@ FastAPI application on port 8001 with health check at `GET /health`. Set `GATEWA
 | **HTML Page Favorites** (`/api/html-page-favorites`) | Bookmarked generated HTML pages, owner-scoped. `GET /` - list; `POST /` `{page_id, title?, description?, tags?}` - bookmark a `scheduled_html_pages` row (extracts a reusable style/layout summary via GLM5 at favorite time); a page can only be favorited once per user (`DuplicateFavoriteError` ? HTTP 409). `GET/DELETE /{id}` - read/delete (the create-task reference picker deletes favorites here). A favorite can be selected as a style reference when creating a new `html_page` task. |
 | **Public Scheduled Tasks** (`/api/public/scheduled-tasks`) | **Unauthenticated** mirror for embedding. Adds `GET /runs/{rid}/html` and `GET /{task_id}/latest-html` for the `html_page` preview; the run result keeps `result.html` (only file paths are redacted). |
 | **Light Apps** (`/api/light-apps`) | Light-app registry (????? / ????), **global/shared** (no user scoping; `created_by`/`updated_by` audit-only). Two kinds: `window_open` (launched in a new tab from the ???? card grid) and `iframe` (embedded under a parent sidebar menu via `mount_location` and dynamically injected into the sidebar). Both carry `route_params` ? a list of `{key, paramType: login\|theme, source, customValue, themeValue}` descriptors the **frontend** resolves at launch time (login params read from `localStorage.userInfo`; theme params use a literal value) and appends to `url` as query string. `GET /` - list (filters `app_type`/`mount_location`/`enabled_only`/`search`); `GET /{app_id}` - read one (both open to any authenticated user); `POST /`, `PUT /{app_id}`, `DELETE /{app_id}` - **admin only**; iframe apps must carry a `mount_location` (400 otherwise); the unique `name` collides as HTTP 409. Backed by `deerflow.persistence.light_apps` (table `light_apps`, auto-created by `create_all`); store wired in `deps.py` as `app.state.light_app_store`. Tests: `tests/test_light_apps.py`. |
-| **Knowledge** (`/api/knowledge`) | Product knowledge base, **global/shared** (no user isolation; writes record `created_by`/`updated_by` for audit). `POST /ingest/thread/{thread_id}` - sediment one thread into a note (reads messages from the checkpointer, `threads:read` owner-checked); `POST /ingest/threads/batch` - batch-sediment the caller's recent threads (`skip_existing` via `.manifest.json` dedup). **Optional model selection**: both ingest endpoints accept `model_name` (overrides `knowledge.extract_model_name` for that capture; threaded through `KnowledgeService.capture_thread` ? `extract_thread_llm_multi`). **Model-error surfacing**: when the chosen model produces nothing usable (timeout / API error / non-JSON) capture still **falls back to rule-based extraction and lands a draft** (`llm_fallback:true`); the single-thread (non-background) response and the batch response (`llm_fallback` count) carry this so the UI tells the user "???????". The "??????" chat button stays background/fire-and-forget (passes `model_name`, no per-save error toast); the batch-import dialog is synchronous and shows the fallback count. `POST /ingest/file` - **import an uploaded file** (multipart: `file` + optional `title`/`tags`/`status`/`model_name`) into wiki note(s): the file is converted to Markdown (md/txt read directly, pdf/word/ppt/excel via `convert_file_to_markdown`), distilled by the same wiki-capture LLM pipeline as thread capture (`source_type="file"`), and falls back to a verbatim **draft** note when the model returns nothing usable (`llm_fallback:true`); `POST /ingest/search` - sediment search/tool results as a reference note; `POST /notes` - manual note; `GET /notes` (keyword/tag/source_type/status + pagination); `GET/PUT/DELETE /notes/{id}` (DELETE is soft = archive, `?hard=true` removes the vault file); `POST /search` - `mode=keyword\|vector\|hybrid` (vector?keyword fallback when no embedding endpoint); `GET /notes/{id}/versions` - edit history; `GET /duplicates` + `POST /merge` - dedup; `POST /reindex` - rebuild vector index; `GET /graph` - product graph data (notes + entities, graph-blacklist filtered); `GET/POST /graph/blacklist` + `DELETE /graph/blacklist/{id}` - graph node blacklist (labels = entity names / note titles, case-insensitive match; blacklisted nodes and their edges are dropped from `/graph` and the wiki-export artifacts; table `knowledge_graph_blacklist`, auto-created); `POST /export/graph` + `GET /export/files/{graph.json\|graph.html\|graph.graphml\|cypher.txt}` - obsidian-wiki `wiki-export` artifacts (self-contained `graph.html` for iframe); `POST /resolve` `{targets:[?]}` + `GET /notes/{id}/wikilinks` - resolve `[[wikilink]]` targets to a note id / vault page / missing (powers clickable wikilinks); `GET /vault-page?path=` - read a raw vault meta/hub Markdown page (no DB note). `GET/POST /folders` + `PUT/DELETE /folders/{id}` - user-managed directories (notes carry a `folder` path; `GET /notes?folder=` filters a dir + sub-dirs); `GET/POST /extract-templates` + `GET/PUT/DELETE /extract-templates/{id}` - editable extraction-prompt templates (thread/document ingest accept `template_id`; capture/PUT-note accept `folder`). Backed by the `deerflow.knowledge` module (see below). |
+| **Knowledge** (`/api/knowledge`) | Product knowledge base, **global/shared** (no user isolation; writes record `created_by`/`updated_by` for audit). `POST /ingest/thread/{thread_id}` - sediment one thread into a note (reads messages from the checkpointer, `threads:read` owner-checked); `POST /ingest/threads/batch` - batch-sediment the caller's recent threads (`skip_existing` via `.manifest.json` dedup). **Defaults to `background=true`**: the slow per-thread extraction loop is deferred to a FastAPI background task and the endpoint returns `{queued:true, processed:N}` immediately (each thread's transcript is truncated to `max_input_chars` before the LLM call). Running the loop inline for a large batch outlives the nginx gateway timeout (504), which the frontend's `parseResponse` surfaces as a generic "Request failed"; `background=false` keeps the blocking path (full per-thread `created/skipped/failed` tally) for small batches/scripts. Shared loop: `_run_batch_ingest`. **Optional model selection**: both ingest endpoints accept `model_name` (overrides `knowledge.extract_model_name` for that capture; threaded through `KnowledgeService.capture_thread` ? `extract_thread_llm_multi`). **Model-error surfacing**: when the chosen model produces nothing usable (timeout / API error / non-JSON) capture still **falls back to rule-based extraction and lands a draft** (`llm_fallback:true`); the single-thread (non-background) response and the batch response (`llm_fallback` count) carry this so the UI tells the user "???????". The "??????" chat button stays background/fire-and-forget (passes `model_name`, no per-save error toast); the batch-import dialog is synchronous and shows the fallback count. `POST /ingest/file` - **import an uploaded file** (multipart: `file` + optional `title`/`tags`/`status`/`model_name`) into wiki note(s): the file is converted to Markdown (md/txt read directly, pdf/word/ppt/excel via `convert_file_to_markdown`), distilled by the same wiki-capture LLM pipeline as thread capture (`source_type="file"`), and falls back to a verbatim **draft** note when the model returns nothing usable (`llm_fallback:true`); `POST /ingest/search` - sediment search/tool results as a reference note; `POST /notes` - manual note; `GET /notes` (keyword/tag/source_type/status + pagination); `GET/PUT/DELETE /notes/{id}` (DELETE is soft = archive, `?hard=true` removes the vault file); `POST /search` - `mode=keyword\|vector\|hybrid` (vector?keyword fallback when no embedding endpoint); `GET /notes/{id}/versions` - edit history; `GET /duplicates` + `POST /merge` - dedup; `POST /reindex` - rebuild vector index; `GET /graph` - product graph data (notes + entities, graph-blacklist filtered); `GET/POST /graph/blacklist` + `DELETE /graph/blacklist/{id}` - graph node blacklist (labels = entity names / note titles, case-insensitive match; blacklisted nodes and their edges are dropped from `/graph` and the wiki-export artifacts; table `knowledge_graph_blacklist`, auto-created); `POST /export/graph` + `GET /export/files/{graph.json\|graph.html\|graph.graphml\|cypher.txt}` - obsidian-wiki `wiki-export` artifacts (self-contained `graph.html` for iframe); `POST /resolve` `{targets:[?]}` + `GET /notes/{id}/wikilinks` - resolve `[[wikilink]]` targets to a note id / vault page / missing (powers clickable wikilinks); `GET /vault-page?path=` - read a raw vault meta/hub Markdown page (no DB note). `GET/POST /folders` + `PUT/DELETE /folders/{id}` - user-managed directories (notes carry a `folder` path; `GET /notes?folder=` filters a dir + sub-dirs); `GET/POST /extract-templates` + `GET/PUT/DELETE /extract-templates/{id}` - editable extraction-prompt templates (thread/document ingest accept `template_id`; capture/PUT-note accept `folder`). Backed by the `deerflow.knowledge` module (see below). |
 
 Proxied through nginx: `/api/langgraph/*` ? LangGraph, all other `/api/*` ? Gateway.
 
@@ -846,10 +846,20 @@ See [docs/plan_mode_usage.md](docs/plan_mode_usage.md) for details.
 ### Context Summarization
 
 Automatic conversation summarization when approaching token limits:
-- Configured in `config.yaml` under `summarization` key
-- Trigger types: tokens, messages, or fraction of max input
-- Keeps recent messages while summarizing older ones
+- **On/off switch is per-request, not the config file.** The chat input box has a
+  「上下文压缩」 toggle (below 参考文献, **default off**) → frontend sends
+  `context.summarization_enabled` → whitelisted in `_CONTEXT_CONFIGURABLE_KEYS`
+  (`app/gateway/services.py`) → `_build_middlewares` reads it from the run config
+  and calls `_create_summarization_middleware(force_enabled=summarization_enabled)`.
+  `force_enabled=True` mounts the middleware even when `config.summarization.enabled`
+  is false; when the flag is absent (channels / scheduled runs) it falls back to
+  the config flag. `config.summarization.enabled` now defaults to **false**.
+- `config.yaml` `summarization` still supplies the **tuning**: trigger types
+  (tokens, messages, or fraction of max input) and the keep policy.
+- Keeps recent messages while summarizing older ones.
 
+Tests: `tests/test_summarization_toggle.py` (gating + whitelist),
+`tests/test_summarization_middleware.py` (cache behavior).
 See [docs/summarization.md](docs/summarization.md) for details.
 
 ### Vision Support / Image Recognition (??)

offline-backend-20260512/backend/app/gateway/routers/knowledge.py

@@ -65,6 +65,10 @@ class BatchIngestRequest(BaseModel):
     include_sources: bool = True
     skip_existing: bool = True
     mode: str = "summary"
+    background: bool = Field(
+        default=True,
+        description="When true (default), the per-thread extraction loop runs in a background task and the endpoint returns immediately ({queued:true}). Batch capture of many threads is far too slow to finish inside one HTTP request — running it inline makes nginx time out (504), which the frontend surfaces as a generic 'request failed'.",
+    )
     model_name: str | None = Field(default=None, description="可选:指定用于知识提炼的大模型名称;留空使用默认配置")
     template_id: str | None = Field(default=None, description="可选:抽取提示词模板 id;留空使用对话默认模板")
     folder: str | None = Field(default=None, description="可选:沉淀结果归入的目录(斜杠分隔路径)")
@@ -285,15 +289,14 @@ async def ingest_thread(thread_id: str, body: IngestThreadRequest, request: Requ
     }
 
 
-@router.post("/ingest/threads/batch")
-@require_auth
-async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> dict[str, Any]:
-    """Batch-sediment the caller's recent threads into knowledge notes."""
-    service = get_knowledge_service(request)
-    checkpointer = get_checkpointer(request)
-    thread_store = get_thread_store(request)
+async def _run_batch_ingest(service, checkpointer, threads: list[dict[str, Any]], body: BatchIngestRequest, created_by: str | None) -> dict[str, Any]:
+    """Run the per-thread sediment loop and tally the outcome.
 
-    threads = await thread_store.search(limit=body.limit, offset=body.offset)
+    Shared by the inline and background paths; ``created_by`` is passed in
+    explicitly because the user contextvar is gone once a background task runs.
+    Each thread's transcript is truncated by the extractor (``max_input_chars``)
+    before the LLM call, so a single huge conversation can't blow the context.
+    """
     created = skipped = failed = 0
     # Notes that landed via the rule-based fallback because the chosen model
     # produced nothing usable (timeout / API error / non-JSON). Surfaced so the
@@ -319,7 +322,7 @@ async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> di
                 status=body.status,
                 include_sources=body.include_sources,
                 reference_batches=batches,
-                created_by=get_effective_user_id(),
+                created_by=created_by,
                 skip_existing=body.skip_existing,
                 model_name=body.model_name,
                 template_id=body.template_id,
@@ -347,6 +350,49 @@ async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> di
     return {"ok": True, "processed": len(items), "created": created, "skipped": skipped, "failed": failed, "llm_fallback": llm_fallback, "items": items}
 
 
+async def _ingest_threads_batch_background(service, checkpointer, threads: list[dict[str, Any]], body: BatchIngestRequest, created_by: str | None) -> None:
+    """Background variant of the batch loop (failures swallowed/logged)."""
+    try:
+        result = await _run_batch_ingest(service, checkpointer, threads, body, created_by)
+        logger.info(
+            "Background batch ingest finished: processed=%d created=%d skipped=%d failed=%d llm_fallback=%d",
+            result["processed"], result["created"], result["skipped"], result["failed"], result["llm_fallback"],
+        )
+    except Exception:
+        logger.exception("Background batch ingest failed")
+
+
+@router.post("/ingest/threads/batch")
+@require_auth
+async def ingest_threads_batch(body: BatchIngestRequest, request: Request, background_tasks: BackgroundTasks) -> dict[str, Any]:
+    """Batch-sediment the caller's recent threads into knowledge notes.
+
+    Defaults to ``background=true``: the (slow) per-thread extraction runs in a
+    background task and the endpoint returns ``{"queued": true, "processed": N}``
+    immediately, so the request never outlives the nginx gateway timeout. Pass
+    ``background=false`` to block until every thread is processed (small batches /
+    scripts that want the full per-thread tally).
+    """
+    service = get_knowledge_service(request)
+    checkpointer = get_checkpointer(request)
+    thread_store = get_thread_store(request)
+
+    threads = await thread_store.search(limit=body.limit, offset=body.offset)
+    if body.background:
+        # Capture the user id now — the contextvar is gone once the task runs.
+        background_tasks.add_task(
+            _ingest_threads_batch_background,
+            service,
+            checkpointer,
+            threads,
+            body,
+            get_effective_user_id(),
+        )
+        return {"ok": True, "queued": True, "processed": len(threads), "created": 0, "skipped": 0, "failed": 0, "llm_fallback": 0, "items": []}
+
+    return await _run_batch_ingest(service, checkpointer, threads, body, get_effective_user_id())
+
+
 async def _extract_upload_text(file: UploadFile) -> tuple[str, str]:
     """Read an uploaded file and return ``(filename, markdown_text)``.
 
@@ -476,7 +522,7 @@ async def list_notes(
     offset: int = 0,
 ) -> dict[str, Any]:
     service = get_knowledge_service(request)
-    limit = max(1, min(limit, 100))
+    limit = max(1, min(limit, 1000))
     offset = max(0, offset)
     return await service.list_notes(keyword=keyword, tag=tag, entity=entity, source_type=source_type, status=status, folder=folder, limit=limit, offset=offset)
 

offline-backend-20260512/backend/app/gateway/services.py

@@ -243,6 +243,9 @@ _CONTEXT_CONFIGURABLE_KEYS: frozenset[str] = frozenset(
         # 关掉个人记忆召回 / builtin 记忆注入(见 multi_agent._run_payload)。
         "memory_recall_disabled",
         "memory_injection_enabled",
+        # 上下文压缩开关:由聊天框下方的按钮控制(默认关闭),透传到 run config 后
+        # 在 _build_middlewares 里决定是否挂载 SummarizationMiddleware。
+        "summarization_enabled",
     }
 )
 

offline-backend-20260512/backend/config.yaml

@@ -200,7 +200,10 @@ title:
 # grep「上下文压缩」/「context compaction」确认生效;确认后再调回生产值
 # (trigger tokens 60000 / messages 80,keep tokens 16000)。
 summarization:
-  enabled: true
+  # 开关已移到聊天框下方(参考文献下面)的「上下文压缩」按钮,默认关闭。
+  # 这里 enabled 仅作为非 Web 路径(渠道/定时任务)的后备开关;下面的 trigger/keep
+  # 等仅是调参,Web 端是否压缩由前端按钮(summarization_enabled)决定。
+  enabled: false
   # model_name: deepseek-chat   # 留空则用主对话模型
   trigger:
     - type: tokens

offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py

@@ -54,12 +54,19 @@ def _resolve_model_name(requested_model_name: str | None = None, *, app_config:
     return default_model_name
 
 
-def _create_summarization_middleware(*, app_config: AppConfig | None = None) -> DeerFlowSummarizationMiddleware | None:
-    """Create and configure the summarization middleware from config."""
+def _create_summarization_middleware(*, app_config: AppConfig | None = None, force_enabled: bool = False) -> DeerFlowSummarizationMiddleware | None:
+    """Create and configure the summarization middleware from config.
+
+    Args:
+        force_enabled: When True, build the middleware regardless of the config
+            ``summarization.enabled`` flag. The per-request 上下文压缩 toggle
+            (chat input box, default off) drives this — the config file no longer
+            owns the on/off switch, only the tuning (trigger/keep/...).
+    """
     resolved_app_config = app_config or get_app_config()
     config = resolved_app_config.summarization
 
-    if not config.enabled:
+    if not (config.enabled or force_enabled):
         return None
 
     # Prepare trigger parameter
@@ -268,8 +275,13 @@ def _build_middlewares(
     if cfg.get("rag_mode_enabled", False):
         middlewares.append(RagCitationMiddleware())
 
-    # Add summarization middleware if enabled
-    summarization_middleware = _create_summarization_middleware(app_config=resolved_app_config)
+    # Add summarization middleware. The on/off switch now lives in the chat UI
+    # (上下文压缩 toggle, default off) and arrives as ``summarization_enabled`` in
+    # the runtime config; the config file only supplies tuning. force_enabled
+    # lets the per-request toggle turn it on even when config.enabled is false,
+    # while non-UI paths (channels/scheduled) still fall back to config.enabled.
+    summarization_requested = cfg.get("summarization_enabled", False)
+    summarization_middleware = _create_summarization_middleware(app_config=resolved_app_config, force_enabled=summarization_requested)
     if summarization_middleware is not None:
         middlewares.append(summarization_middleware)
 

offline-backend-20260512/config.yaml

@@ -303,7 +303,10 @@ log_system:
   label: "日志系统"
 
 summarization:
-  enabled: true
+  # 开关已移到聊天框下方(参考文献下面)的「上下文压缩」按钮,默认关闭。
+  # 这里 enabled 仅作为非 Web 路径(渠道/定时任务)的后备开关;下面的 trigger/keep
+  # 等仅是调参,Web 端是否压缩由前端按钮(summarization_enabled)决定。
+  enabled: false
   # model_name: deepseek-chat
   trigger:
     - type: tokens