diff --git a/AGENTS.md b/AGENTS.md
index 3272a45c..e2481373 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -134,7 +134,7 @@ A change is done only when:
 ## Current Stats (v0.9.24)
 
 - 55 MCP tools (8 visible by default, `AGENTMEMORY_TOOLS=all` for all)
-- 141 REST endpoints
+- 142 REST endpoints
 - 6 MCP resources, 3 MCP prompts
 - 12 hooks, 9 skills
 - 60+ iii functions
diff --git a/README.md b/README.md
index aaf4d959..47d26d22 100644
--- a/README.md
+++ b/README.md
@@ -74,7 +74,7 @@ LLM extraction also writes classification tags such as `type:waiting_for_user`,
 
 ## Under the hood
 
-Under the hood, the current prototype still exposes the full implementation surface: **55 MCP tools** (8 visible by default — 55 tools, 6 resources, 3 prompts over MCP) and a local REST API serving **141 endpoints on port** 3111. These counts track the implementation that is mid-rename to AI Todo.
+Under the hood, the current prototype still exposes the full implementation surface: **55 MCP tools** (8 visible by default — 55 tools, 6 resources, 3 prompts over MCP) and a local REST API serving **142 endpoints on port** 3111. These counts track the implementation that is mid-rename to AI Todo.
 
 ## Documentation
 
diff --git a/src/config.ts b/src/config.ts
index ac94a421..20730043 100644
--- a/src/config.ts
+++ b/src/config.ts
@@ -22,18 +22,15 @@ export const DEFAULT_LANGEXTRACT_MODEL = "deepseek/deepseek-v4-flash";
 export const DEFAULT_LANGEXTRACT_PROVIDER = "openai";
 export const DEFAULT_LANGEXTRACT_BASE_URL = "https://api.novita.ai/openai/v1";
 export const DEFAULT_TODO_EXTRACT_TIMEOUT_MS = 120_000;
+export const DEFAULT_TODO_EXTRACT_MAX_LLM_SESSIONS = 12;
 // STEP-11: extraction scope. sinceDays = only sessions whose endedAt/startedAt
 // falls within the last N days are eligible (the primary scope control). Max
 // interactions = per session, keep at most M most-recent interaction records
 // (a "turn": one user message through everything before the next). maxSessions
-// stays a backend safety cap (not surfaced as a setting).
+// has no default cap; REST callers can still pass an explicit positive cap.
 export const DEFAULT_TODO_EXTRACT_SINCE_DAYS = 7;
 export const DEFAULT_TODO_EXTRACT_MAX_INTERACTIONS = 10;
-// Interactive safety cap on how many sessions one extraction pass touches. The
-// day window is the primary control, but with the LLM extractor each session is
-// a serial sidecar call (up to the per-call timeout), so a single "organize"
-// click must stay bounded. REST callers can still pass a larger maxSessions.
-export const DEFAULT_TODO_EXTRACT_MAX_SESSIONS = 8;
+export const DEFAULT_TODO_EXTRACT_MAX_SESSIONS = Number.POSITIVE_INFINITY;
 const LEGACY_LANGEXTRACT_MODELS = new Set(["pa/gpt-5.5"]);
 export const WRITABLE_TODO_EXTRACT_KEYS = new Set([
   "AGENTMEMORY_TODO_EXTRACTOR",
@@ -46,6 +43,7 @@ export const WRITABLE_TODO_EXTRACT_KEYS = new Set([
   "AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS",
   "AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS",
   "AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION",
+  "AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS",
 ]);
 
 let warnPremiumModelShown = false;
@@ -96,6 +94,8 @@ export function getTodoExtractorUserConfig(): Record<string, string | boolean> {
     LANGEXTRACT_THINKING_DEPTH: env["LANGEXTRACT_THINKING_DEPTH"] || "medium",
     AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS:
       env["AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS"] || String(DEFAULT_TODO_EXTRACT_TIMEOUT_MS),
+    AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS:
+      env["AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS"] || String(DEFAULT_TODO_EXTRACT_MAX_LLM_SESSIONS),
     AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS:
       env["AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS"] || String(DEFAULT_TODO_EXTRACT_SINCE_DAYS),
     AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION:
diff --git a/src/functions/todo-extract.ts b/src/functions/todo-extract.ts
index 3983bfd6..88bcb305 100644
--- a/src/functions/todo-extract.ts
+++ b/src/functions/todo-extract.ts
@@ -8,6 +8,7 @@ import { KV, fingerprintId, generateId, nearDuplicateTitle } from "../state/sche
 import type { Action, CompressedObservation, ReviewQueueItem, ScanCheckpoint, Session } from "../types.js";
 import {
   DEFAULT_LANGEXTRACT_BASE_URL,
+  DEFAULT_TODO_EXTRACT_MAX_LLM_SESSIONS,
   DEFAULT_TODO_EXTRACT_TIMEOUT_MS,
   DEFAULT_TODO_EXTRACT_SINCE_DAYS,
   DEFAULT_TODO_EXTRACT_MAX_INTERACTIONS,
@@ -59,6 +60,7 @@ type TodoExtractOptions = {
   // then the config defaults when omitted.
   sinceDays?: number;
   maxInteractionsPerSession?: number;
+  maxLlmSessions?: number;
   project?: string;
   force?: boolean;
   scanSources?: boolean;
@@ -81,14 +83,78 @@ const SIDE_CAR_ENV_KEYS = [
   "LANGEXTRACT_MAX_CHAR_BUFFER",
 ];
 
+export type TodoExtractErrorCode =
+  | "llm_unavailable"
+  | "provider_timeout"
+  | "config_error"
+  | "provider_error"
+  | "extract_failed";
+
+export type TodoExtractJobStatus = "idle" | "running" | "done" | "error";
+
+export type TodoExtractResult = {
+  success: true;
+  jobId?: string;
+  status?: TodoExtractJobStatus;
+  startedAt?: string;
+  finishedAt?: string;
+  engine: "langextract" | "rules" | "mixed";
+  scannedSessions: number;
+  processedSessions: number;
+  skippedUnchangedSessions: number;
+  scannedObservations: number;
+  directCreated: number;
+  reviewCreated: number;
+  hiddenHistory: number;
+  discarded: number;
+  cleanedActions: number;
+  cleanedReviews: number;
+  completedActions: number;
+  completedReviews: number;
+  recheckMarked: number;
+  llmSessionBudget: number;
+  llmSessionsAttempted: number;
+  llmSessionsSkipped: number;
+  llmFallback?: boolean;
+  fallbackReason?: string;
+  errorCode?: TodoExtractErrorCode;
+  cleanupPreview?: { actions: unknown[]; reviews: unknown[] };
+  sourceScan?: { imported: number; skipped: number; errors: number };
+};
+
+export type TodoExtractJob = {
+  success: boolean;
+  jobId: string;
+  status: TodoExtractJobStatus;
+  startedAt: string;
+  finishedAt?: string;
+  message?: string;
+  errorCode?: TodoExtractErrorCode;
+  errorMessage?: string;
+  result?: TodoExtractResult;
+  inFlight?: boolean;
+};
+
+let activeTodoExtractJob: Promise<TodoExtractResult> | null = null;
+let currentTodoExtractJob: TodoExtractJob | null = null;
+
 function envNumber(key: string, fallback: number): number {
   const parsed = Number(getEnvVar(key));
   return Number.isFinite(parsed) ? parsed : fallback;
 }
 
+function classifyExtractError(message: string | undefined): TodoExtractErrorCode {
+  const text = String(message || "").toLowerCase();
+  if (/(timed out|timeout|abort)/.test(text)) return "provider_timeout";
+  if (/(api key|apikey|unauthorized|forbidden|401|403|required)/.test(text)) return "config_error";
+  if (/(rate limit|429|quota|provider|openai|novita|langextract)/.test(text)) return "provider_error";
+  return "extract_failed";
+}
+
 function clampPositiveInt(value: unknown, fallback: number, max: number): number {
   const parsed = typeof value === "number" ? value : parseInt(String(value ?? ""), 10);
   if (!Number.isFinite(parsed) || parsed < 1) return fallback;
+  if (!Number.isFinite(max)) return Math.floor(parsed);
   return Math.min(max, Math.floor(parsed));
 }
 
@@ -1134,12 +1200,13 @@ async function extractForSession(
   session: Session,
   observations: CompressedObservation[],
   mode: string,
-): Promise<{ todos: ExtractedTodo[]; engine: "langextract" | "rules"; fallbackReason?: string }> {
+  opts: { allowLlm?: boolean } = {},
+): Promise<{ todos: ExtractedTodo[]; engine: "langextract" | "rules"; fallbackReason?: string; errorCode?: TodoExtractErrorCode }> {
   const { ruleObservations, llmObservations } = prefilterTodoObservations(session, observations);
   const blocks = llmObservations.map(blockFor).filter((block) => block.text);
   const bucket = timeBucketFor(session);
   let fallbackReason = "";
-  if (mode !== "rules" && blocks.length > 0) {
+  if (mode !== "rules" && opts.allowLlm !== false && blocks.length > 0) {
     try {
       const todos = await runLangExtractSidecar({
         sessionId: session.id,
@@ -1160,31 +1227,14 @@ async function extractForSession(
     todos: candidates.map((candidate) => candidateToTodo(candidate, session, bucket)).filter((todo): todo is ExtractedTodo => !!todo),
     engine: "rules",
     ...(fallbackReason ? { fallbackReason } : {}),
+    ...(fallbackReason ? { errorCode: classifyExtractError(fallbackReason) } : {}),
   };
 }
 
 export async function generateTodosFromSessions(
   kv: Pick<StateKV, "get" | "set" | "list" | "delete">,
   data: TodoExtractOptions = {},
-): Promise<{
-  success: true;
-  engine: "langextract" | "rules" | "mixed";
-  scannedSessions: number;
-  scannedObservations: number;
-  directCreated: number;
-  reviewCreated: number;
-  hiddenHistory: number;
-  discarded: number;
-  cleanedActions: number;
-  cleanedReviews: number;
-  completedActions: number;
-  completedReviews: number;
-  cleanupPreview?: { actions: CleanupPreviewItem[]; reviews: CleanupPreviewItem[] };
-  recheckMarked: number;
-  sourceScan?: { imported: number; skipped: number; errors: number };
-  llmFallback?: boolean;
-  fallbackReason?: string;
-}> {
+): Promise<TodoExtractResult> {
   let sourceScan: { imported: number; skipped: number; errors: number } | undefined;
   if (data.scanSources !== false) {
     const scan = await scanCodexSource(kv as StateKV).catch(() => null);
@@ -1204,6 +1254,11 @@ export async function generateTodosFromSessions(
     DEFAULT_TODO_EXTRACT_MAX_INTERACTIONS,
     500,
   );
+  const maxLlmSessions = clampPositiveInt(
+    data.maxLlmSessions ?? getEnvVar("AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS"),
+    DEFAULT_TODO_EXTRACT_MAX_LLM_SESSIONS,
+    100,
+  );
   const sinceCutoffMs = Date.now() - sinceDays * 24 * 60 * 60 * 1000;
   const mode = (getEnvVar("AGENTMEMORY_TODO_EXTRACTOR") || "auto").toLowerCase();
   const directThreshold = envNumber("AGENTMEMORY_TODO_DIRECT_CONFIDENCE", 0.82);
@@ -1233,8 +1288,8 @@ export async function generateTodosFromSessions(
   const sessions = allSessions
     .filter((session) => !data.project || session.project === data.project || session.cwd === data.project)
     // STEP-11: day-window is the primary scope control. Sessions with no/invalid
-    // timestamp are kept (never silently drop work); maxSessions is the cap that
-    // still bounds a day with a flood of sessions.
+    // timestamp are kept (never silently drop work); maxSessions only applies
+    // when a caller explicitly passes a positive cap.
     .filter((session) => {
       const raw = sessionSortTime(session);
       if (!raw) return true;
@@ -1248,14 +1303,22 @@ export async function generateTodosFromSessions(
   let reviewCreated = 0;
   let hiddenHistory = 0;
   let discarded = 0;
+  let processedSessions = 0;
+  let skippedUnchangedSessions = 0;
+  let llmSessionsAttempted = 0;
+  let llmSessionsSkipped = 0;
   const engines = new Set<"langextract" | "rules">();
   const fallbackReasons = new Set<string>();
+  const errorCodes = new Set<TodoExtractErrorCode>();
   const now = new Date().toISOString();
   const recheckMarked = await markChangedGeneratedActions(kv, new Map(allSessions.map((session) => [session.id, session])), now);
 
   for (const session of sessions) {
     const key = checkpointKey(session);
-    if (!data.force && processed[session.id] === key) continue;
+    if (!data.force && processed[session.id] === key) {
+      skippedUnchangedSessions++;
+      continue;
+    }
     const sortedObservations = (await kv.list<CompressedObservation>(KV.observations(session.id)).catch(() => []))
       .sort((a, b) => (a.timestamp || "").localeCompare(b.timestamp || ""));
     // STEP-11: keep only the most recent N interaction records, then apply the
@@ -1266,9 +1329,14 @@ export async function generateTodosFromSessions(
     scannedObservations += ruleObservations.length;
     const evidenceObservations = mode === "rules" ? ruleObservations : llmObservations.length ? llmObservations : ruleObservations;
     const blockMap = new Map(evidenceObservations.map((obs) => [obs.id, blockFor(obs)]));
-    const { todos, engine, fallbackReason } = await extractForSession(session, ruleObservations, mode);
+    const wantsLlm = mode !== "rules" && llmObservations.length > 0 && timeBucketFor(session) !== "history";
+    const allowLlm = !wantsLlm || llmSessionsAttempted < maxLlmSessions;
+    if (wantsLlm && allowLlm) llmSessionsAttempted++;
+    if (wantsLlm && !allowLlm) llmSessionsSkipped++;
+    const { todos, engine, fallbackReason, errorCode } = await extractForSession(session, ruleObservations, mode, { allowLlm });
     engines.add(engine);
     if (fallbackReason) fallbackReasons.add(fallbackReason.slice(0, 240));
+    if (errorCode) errorCodes.add(errorCode);
     for (const rawTodo of todos) {
       const todo = todoForStorage(rawTodo);
       if (!todo || !validateTodoEvidence(todo, blockMap)) {
@@ -1318,6 +1386,7 @@ export async function generateTodosFromSessions(
       seenTitles.push(titleKey);
     }
     processed[session.id] = key;
+    processedSessions++;
   }
 
   await kv.set(KV.scanCheckpoints, checkpointId, {
@@ -1330,6 +1399,8 @@ export async function generateTodosFromSessions(
     success: true,
     engine: engines.size > 1 ? "mixed" : Array.from(engines)[0] || "rules",
     scannedSessions: sessions.length,
+    processedSessions,
+    skippedUnchangedSessions,
     scannedObservations,
     directCreated,
     reviewCreated,
@@ -1340,17 +1411,122 @@ export async function generateTodosFromSessions(
     completedActions: cleanup.completedActions,
     completedReviews: cleanup.completedReviews,
     recheckMarked,
+    llmSessionBudget: maxLlmSessions,
+    llmSessionsAttempted,
+    llmSessionsSkipped,
     ...(data.cleanup === "dry-run" ? { cleanupPreview: cleanup.preview } : {}),
     ...(sourceScan ? { sourceScan } : {}),
     ...(fallbackReasons.size ? { llmFallback: true } : {}),
     ...(fallbackReasons.size ? { fallbackReason: Array.from(fallbackReasons)[0] } : {}),
+    ...(errorCodes.size ? { errorCode: Array.from(errorCodes)[0] } : {}),
   };
 }
 
+export function getTodoExtractJobStatus(): TodoExtractJob {
+  if (!currentTodoExtractJob) {
+    return {
+      success: true,
+      jobId: "",
+      status: "idle",
+      startedAt: "",
+    };
+  }
+  return { ...currentTodoExtractJob };
+}
+
+export async function startTodoExtractJob(
+  kv: Pick<StateKV, "get" | "set" | "list" | "delete">,
+  data: TodoExtractOptions = {},
+): Promise<TodoExtractJob> {
+  if (activeTodoExtractJob && currentTodoExtractJob?.status === "running") {
+    return { ...currentTodoExtractJob, success: true, inFlight: true };
+  }
+
+  const jobId = generateId("todo_extract");
+  const startedAt = new Date().toISOString();
+  currentTodoExtractJob = {
+    success: true,
+    jobId,
+    status: "running",
+    startedAt,
+    message: "running",
+    inFlight: true,
+  };
+
+  activeTodoExtractJob = generateTodosFromSessions(kv, data)
+    .then((result) => {
+      const finishedAt = new Date().toISOString();
+      const next: TodoExtractResult = {
+        ...result,
+        jobId,
+        status: "done",
+        startedAt,
+        finishedAt,
+      };
+      currentTodoExtractJob = {
+        success: true,
+        jobId,
+        status: "done",
+        startedAt,
+        finishedAt,
+        result: next,
+      };
+      return next;
+    })
+    .catch((err) => {
+      const message = err instanceof Error ? err.message : String(err || "todo extraction failed");
+      const errorCode = classifyExtractError(message);
+      currentTodoExtractJob = {
+        success: false,
+        jobId,
+        status: "error",
+        startedAt,
+        finishedAt: new Date().toISOString(),
+        errorCode,
+        errorMessage: message,
+      };
+      throw err;
+    })
+    .finally(() => {
+      activeTodoExtractJob = null;
+    });
+
+  return { ...currentTodoExtractJob, inFlight: false };
+}
+
+function isDuplicateTodoExtractJob(job: TodoExtractJob): boolean {
+  return job.inFlight === true;
+}
+
+export async function runTodoExtractJob(
+  kv: Pick<StateKV, "get" | "set" | "list" | "delete">,
+  data: TodoExtractOptions = {},
+): Promise<TodoExtractJob> {
+  const job = await startTodoExtractJob(kv, data);
+  if (isDuplicateTodoExtractJob(job) && job.status === "running") {
+    return job;
+  }
+  if (!activeTodoExtractJob) return getTodoExtractJobStatus();
+
+  try {
+    const result = await activeTodoExtractJob;
+    return {
+      ...result,
+      success: true,
+      jobId: result.jobId || job.jobId,
+      status: "done",
+      startedAt: result.startedAt || job.startedAt,
+      finishedAt: result.finishedAt,
+      result,
+    } as TodoExtractJob;
+  } catch {
+    return getTodoExtractJobStatus();
+  }
+}
+
 export function registerTodoExtractFunctions(sdk: ISdk, kv: StateKV): void {
-  sdk.registerFunction("mem::todo-extract-generate", async (data: TodoExtractOptions = {}) =>
-    generateTodosFromSessions(kv, data),
-  );
+  sdk.registerFunction("mem::todo-extract-generate", async (data: TodoExtractOptions = {}) => runTodoExtractJob(kv, data));
+  sdk.registerFunction("mem::todo-extract-status", async () => getTodoExtractJobStatus());
   sdk.registerFunction("mem::todo-update", async (data: TodoUpdateOptions = {}) =>
     updateChangedTodoCards(kv, data),
   );
diff --git a/src/index.ts b/src/index.ts
index 91fd8852..3aa47834 100644
--- a/src/index.ts
+++ b/src/index.ts
@@ -531,7 +531,7 @@ async function main() {
     `Ready. ${embeddingProvider ? "Triple-stream (BM25+Vector+Graph)" : "BM25+Graph"} search active.`,
   );
   bootLog(
-    `REST API: 141 endpoints at http://localhost:${config.restPort}/agentmemory/*`,
+    `REST API: 142 endpoints at http://localhost:${config.restPort}/agentmemory/*`,
   );
   bootLog(
     `MCP surface (opt-in via \`npx @agentmemory/mcp\`): ${getAllTools().length} tools · 6 resources · 3 prompts`,
diff --git a/src/triggers/api.ts b/src/triggers/api.ts
index 7fe3cb84..3f0ec849 100644
--- a/src/triggers/api.ts
+++ b/src/triggers/api.ts
@@ -522,7 +522,7 @@ export function registerApiTriggers(
           success: true,
           envPath: getUserEnvPath(),
           config: getTodoExtractorUserConfig(),
-          restartRequired: true,
+          restartRequired: false,
         },
       };
     },
@@ -1510,11 +1510,16 @@ export function registerApiTriggers(
       if (maxInteractionsPerSession === null) {
         return { status_code: 400, body: { error: "maxInteractionsPerSession must be a positive integer" } };
       }
+      const maxLlmSessions = parseOptionalPositiveInt(body.maxLlmSessions);
+      if (maxLlmSessions === null) {
+        return { status_code: 400, body: { error: "maxLlmSessions must be a positive integer" } };
+      }
       const payload: Record<string, unknown> = {};
       if (maxSessions !== undefined) payload.maxSessions = maxSessions;
       if (maxObservationsPerSession !== undefined) payload.maxObservationsPerSession = maxObservationsPerSession;
       if (sinceDays !== undefined) payload.sinceDays = sinceDays;
       if (maxInteractionsPerSession !== undefined) payload.maxInteractionsPerSession = maxInteractionsPerSession;
+      if (maxLlmSessions !== undefined) payload.maxLlmSessions = maxLlmSessions;
       const project = asNonEmptyString(body.project);
       if (project) payload.project = project;
       if (body.force === true) payload.force = true;
@@ -1528,6 +1533,19 @@ export function registerApiTriggers(
     function_id: "api::todo-extract-generate",
     config: { api_path: "/agentmemory/todo-extract/generate", http_method: "POST" },
   });
+  sdk.registerFunction("api::todo-extract-status",
+    async (req: ApiRequest): Promise<Response> => {
+      const authErr = checkAuth(req, secret);
+      if (authErr) return authErr;
+      const result = await sdk.trigger({ function_id: "mem::todo-extract-status", payload: {} });
+      return { status_code: 200, body: result };
+    },
+  );
+  sdk.registerTrigger({
+    type: "http",
+    function_id: "api::todo-extract-status",
+    config: { api_path: "/agentmemory/todo-extract/status", http_method: "GET" },
+  });
 
   sdk.registerFunction("api::todo-update",
     async (req: ApiRequest): Promise<Response> => {
diff --git a/src/viewer/index.html b/src/viewer/index.html
index ee768b00..44460812 100644
--- a/src/viewer/index.html
+++ b/src/viewer/index.html
@@ -4163,6 +4163,10 @@ <h1>AI Todo</h1>
         'act.extract.rules': 'LLM unavailable',
         'act.extract.error': 'Organize failed',
         'act.extract.failedExisting': 'Extraction failed; showing existing todos',
+        'act.extract.timeout': 'Provider timed out; showing existing todos',
+        'act.extract.configError': 'LLM config needs attention; showing existing todos',
+        'act.extract.providerError': 'LLM provider unavailable; showing existing todos',
+        'act.extract.runningExisting': 'Still organizing from a previous request...',
         'act.extract.loading': 'Loading todos...',
         'act.extract.starting': 'Organizing recent sessions...',
         'act.extract.background': 'Latest todos are shown; still organizing...',
@@ -4192,10 +4196,11 @@ <h1>AI Todo</h1>
         'act.empty.title': 'No todos yet',
         'act.empty.lead': 'This is where todos, blocked items, and completed work extracted from your sessions will appear.',
         'settings.title': 'Settings',
-        'settings.subtitle': 'Local configuration is written to the user config file and takes effect after restarting the service.',
+        'settings.subtitle': 'Local configuration is written to the user config file and applies to the next organize run.',
         'settings.close': 'Close',
         'settings.language': 'UI language',
         'settings.extractor': 'LLM extraction config',
+        'settings.maxLlmSessions': 'Max LLM sessions per organize run',
         'settings.sinceDays': 'Look-back window (days): only sessions from the last N days',
         'settings.maxInteractions': 'Max interaction records per session (one user request → agent reply)',
         'settings.apiKeyKeep': 'Enter a new API key to replace it, or leave blank to keep the current key',
@@ -4203,7 +4208,7 @@ <h1>AI Todo</h1>
         'settings.apiKeyLabel': 'API key:',
         'settings.save': 'Save config',
         'settings.saving': 'Saving...',
-        'settings.savedRestart': 'Config saved. Restart the service to apply it.',
+        'settings.savedRestart': 'Config saved. It applies to the next organize run.',
         'settings.saveFailed': 'Config save failed',
         'act.status.updateFailed': 'Todo status update failed',
         'obs.type.file_read': 'Read file',
@@ -4365,6 +4370,10 @@ <h1>AI Todo</h1>
         'act.extract.rules': '未走大模型',
         'act.extract.error': '整理失败',
         'act.extract.failedExisting': '抽取失败，已显示现有待办',
+        'act.extract.timeout': '上游超时，已显示现有待办',
+        'act.extract.configError': '大模型配置需要检查，已显示现有待办',
+        'act.extract.providerError': '大模型服务不可用，已显示现有待办',
+        'act.extract.runningExisting': '上一次整理仍在进行...',
         'act.extract.loading': '正在整理待办...',
         'act.extract.starting': '正在从最近会话整理待办...',
         'act.extract.background': '已显示最新待办，后台仍在整理...',
@@ -4394,10 +4403,11 @@ <h1>AI Todo</h1>
         'act.empty.title': '还没有待办',
         'act.empty.lead': '这里会放从会话里整理出的待办、卡住事项和已完成事项。',
         'settings.title': '设置',
-        'settings.subtitle': '本机配置会写入用户配置文件，重启服务后生效。',
+        'settings.subtitle': '本机配置会写入用户配置文件，下次整理时生效。',
         'settings.close': '关闭',
         'settings.language': '界面语言',
         'settings.extractor': '大模型抽取配置',
+        'settings.maxLlmSessions': '每次整理最多调用大模型的会话数',
         'settings.sinceDays': '回溯天数：只抽取最近 N 天内的会话',
         'settings.maxInteractions': '每会话最多交互记录数（一次用户派发→Agent 回复为一条）',
         'settings.apiKeyKeep': '输入新 API key 覆盖，留空保持不变',
@@ -4405,7 +4415,7 @@ <h1>AI Todo</h1>
         'settings.apiKeyLabel': 'API key:',
         'settings.save': '保存配置',
         'settings.saving': '保存中...',
-        'settings.savedRestart': '配置已保存，重启后生效。',
+        'settings.savedRestart': '配置已保存，下次整理时生效。',
         'settings.saveFailed': '配置保存失败',
         'act.status.updateFailed': '待办状态更新失败',
         'obs.type.file_read': '读取文件',
@@ -4614,7 +4624,7 @@ <h1>AI Todo</h1>
       audit: { loaded: false, entries: [], opFilter: '' },
       activity: { loaded: false, observations: [], sessions: [], typeFilter: '', loadingPhase: '', warnings: [] },
       lessons: { loaded: false, items: [], search: '', skillSearch: '', skillRootFilter: 'all', mode: 'explicit', projects: [] },
-      actions: { loaded: false, items: [], reviewItems: [], frontier: [], statusFilter: '', search: '', doneExpanded: false, extractStatus: '', extractMessage: '', extractInFlight: false, stale: false, config: null, configSaving: false, configDraft: {} },
+      actions: { loaded: false, items: [], reviewItems: [], frontier: [], statusFilter: '', search: '', doneExpanded: false, extractStatus: '', extractMessage: '', extractInFlight: false, extractJob: null, stale: false, config: null, configSaving: false, configDraft: {} },
       inbox: { loaded: false, items: [], awaitingItems: [], answeredItems: [], dismissedItems: [], replyingId: null, pendingById: {}, briefingExpanded: false, answeredExpanded: false },
       crystals: { loaded: false, items: [], search: '', lessonMap: {} },
       profile: { loaded: false, projects: [], selectedProject: '', data: null },
@@ -4624,7 +4634,6 @@ <h1>AI Todo</h1>
       settings: { open: false },
       ws: null
     };
-
     function esc(s) {
       if (!s) return '';
       var d = document.createElement('div');
@@ -8720,6 +8729,7 @@ <h1>AI Todo</h1>
       if (state.settings.open) {
         loadTodoExtractorConfig().then(renderSettingsPanel).catch(function() {});
       }
+      apiGet('todo-extract/status').then(syncTodoExtractJob).catch(function() {});
       if (opts.generate === true) startTodoExtraction(opts.force === true);
     }
 
@@ -8746,6 +8756,43 @@ <h1>AI Todo</h1>
       return !!result && (result.engine === 'langextract' || result.engine === 'mixed') && !result.llmFallback;
     }
 
+    function todoExtractionErrorMessage(result) {
+      var code = result && (result.errorCode || (result.result && result.result.errorCode));
+      if (code === 'provider_timeout') return t('act.extract.timeout');
+      if (code === 'config_error') return t('act.extract.configError');
+      if (code === 'provider_error' || code === 'llm_unavailable') return t('act.extract.providerError');
+      return t('act.extract.failedExisting');
+    }
+
+    function todoExtractionResultFromJob(job) {
+      if (!job) return null;
+      return job.result || (job.success === true && job.engine ? job : null);
+    }
+
+    function syncTodoExtractJob(job) {
+      if (!job || !job.status || job.status === 'idle') return job;
+      state.actions.extractJob = job;
+      if (job.status === 'running') {
+        state.actions.extractInFlight = true;
+        state.actions.extractStatus = 'running';
+        state.actions.extractMessage = t('act.extract.runningExisting');
+        if (state.activeTab === 'actions') renderActions();
+        return job;
+      }
+      state.actions.extractInFlight = false;
+      if (job.status === 'done') {
+        var result = todoExtractionResultFromJob(job);
+        state.actions.extractStatus = 'done';
+        state.actions.extractFallback = !todoExtractionUsedLlm(result);
+        state.actions.extractMessage = todoExtractionSummary(result);
+      } else if (job.status === 'error') {
+        state.actions.extractStatus = 'error';
+        state.actions.extractMessage = todoExtractionErrorMessage(job);
+      }
+      if (state.activeTab === 'actions') renderActions();
+      return job;
+    }
+
     function refreshActionListsAfterExtract() {
       return Promise.all([
         apiGet('actions'),
@@ -8817,6 +8864,7 @@ <h1>AI Todo</h1>
       html += '<input id="todo-config-LANGEXTRACT_BASE_URL" class="search-input" value="' + value('LANGEXTRACT_BASE_URL', 'https://api.novita.ai/openai/v1') + '" placeholder="https://api.novita.ai/openai/v1" />';
       html += '<input id="todo-config-LANGEXTRACT_THINKING_DEPTH" class="search-input" value="' + value('LANGEXTRACT_THINKING_DEPTH', 'medium') + '" placeholder="medium" />';
       html += '<input id="todo-config-AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS" class="search-input" value="' + value('AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS', '120000') + '" placeholder="120000" />';
+      html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.maxLlmSessions')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS', '12') + '" placeholder="12" /></div>';
       html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.sinceDays')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS', '7') + '" placeholder="7" /></div>';
       html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.maxInteractions')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION', '10') + '" placeholder="10" /></div>';
       html += '<div><input id="todo-config-LANGEXTRACT_API_KEY" class="search-input" type="password" placeholder="' + esc(keyLabel) + '" />';
@@ -8841,6 +8889,7 @@ <h1>AI Todo</h1>
         'LANGEXTRACT_BASE_URL',
         'LANGEXTRACT_THINKING_DEPTH',
         'AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS',
+        'AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS',
         'AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS',
         'AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION',
         'LANGEXTRACT_API_KEY'
@@ -8861,6 +8910,7 @@ <h1>AI Todo</h1>
         'LANGEXTRACT_BASE_URL',
         'LANGEXTRACT_THINKING_DEPTH',
         'AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS',
+        'AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS',
         'AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS',
         'AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION',
         'LANGEXTRACT_API_KEY'
@@ -8937,13 +8987,21 @@ <h1>AI Todo</h1>
       // settings would never take effect on this primary extraction path.
       apiPost('todo-extract/generate', {
         force: force === true
-      }).then(function(result) {
+      }).then(function(job) {
+        var result = todoExtractionResultFromJob(job);
+        if (job && job.status === 'running') {
+          state.actions.extractJob = job;
+          state.actions.extractStatus = 'running';
+          state.actions.extractMessage = t('act.extract.runningExisting');
+          return refreshActionListsAfterExtract();
+        }
         var delta = todoExtractionDelta(result);
         if (!result || result.success !== true) {
           state.actions.extractStatus = 'error';
-          state.actions.extractMessage = t('act.extract.failedExisting');
+          state.actions.extractMessage = todoExtractionErrorMessage(job || result);
           return null;
         }
+        state.actions.extractJob = job;
         state.actions.extractStatus = 'done';
         state.actions.extractFallback = !todoExtractionUsedLlm(result);
         state.actions.extractMessage = todoExtractionSummary(result);
@@ -8957,7 +9015,9 @@ <h1>AI Todo</h1>
         state.actions.extractMessage = t('act.extract.failedExisting');
       }).then(function() {
         clearTimeout(softRefreshTimer);
-        state.actions.extractInFlight = false;
+        if (!state.actions.extractJob || state.actions.extractJob.status !== 'running') {
+          state.actions.extractInFlight = false;
+        }
         if (state.activeTab === 'actions' && !actionsScrolledAway()) {
           renderActions();
         } else if (state.activeTab !== 'actions') {
diff --git a/src/viewer/parts/app/05-i18n.js b/src/viewer/parts/app/05-i18n.js
index b38934f9..c960732f 100644
--- a/src/viewer/parts/app/05-i18n.js
+++ b/src/viewer/parts/app/05-i18n.js
@@ -63,6 +63,10 @@
         'act.extract.rules': 'LLM unavailable',
         'act.extract.error': 'Organize failed',
         'act.extract.failedExisting': 'Extraction failed; showing existing todos',
+        'act.extract.timeout': 'Provider timed out; showing existing todos',
+        'act.extract.configError': 'LLM config needs attention; showing existing todos',
+        'act.extract.providerError': 'LLM provider unavailable; showing existing todos',
+        'act.extract.runningExisting': 'Still organizing from a previous request...',
         'act.extract.loading': 'Loading todos...',
         'act.extract.starting': 'Organizing recent sessions...',
         'act.extract.background': 'Latest todos are shown; still organizing...',
@@ -92,10 +96,11 @@
         'act.empty.title': 'No todos yet',
         'act.empty.lead': 'This is where todos, blocked items, and completed work extracted from your sessions will appear.',
         'settings.title': 'Settings',
-        'settings.subtitle': 'Local configuration is written to the user config file and takes effect after restarting the service.',
+        'settings.subtitle': 'Local configuration is written to the user config file and applies to the next organize run.',
         'settings.close': 'Close',
         'settings.language': 'UI language',
         'settings.extractor': 'LLM extraction config',
+        'settings.maxLlmSessions': 'Max LLM sessions per organize run',
         'settings.sinceDays': 'Look-back window (days): only sessions from the last N days',
         'settings.maxInteractions': 'Max interaction records per session (one user request → agent reply)',
         'settings.apiKeyKeep': 'Enter a new API key to replace it, or leave blank to keep the current key',
@@ -103,7 +108,7 @@
         'settings.apiKeyLabel': 'API key:',
         'settings.save': 'Save config',
         'settings.saving': 'Saving...',
-        'settings.savedRestart': 'Config saved. Restart the service to apply it.',
+        'settings.savedRestart': 'Config saved. It applies to the next organize run.',
         'settings.saveFailed': 'Config save failed',
         'act.status.updateFailed': 'Todo status update failed',
         'obs.type.file_read': 'Read file',
@@ -265,6 +270,10 @@
         'act.extract.rules': '未走大模型',
         'act.extract.error': '整理失败',
         'act.extract.failedExisting': '抽取失败，已显示现有待办',
+        'act.extract.timeout': '上游超时，已显示现有待办',
+        'act.extract.configError': '大模型配置需要检查，已显示现有待办',
+        'act.extract.providerError': '大模型服务不可用，已显示现有待办',
+        'act.extract.runningExisting': '上一次整理仍在进行...',
         'act.extract.loading': '正在整理待办...',
         'act.extract.starting': '正在从最近会话整理待办...',
         'act.extract.background': '已显示最新待办，后台仍在整理...',
@@ -294,10 +303,11 @@
         'act.empty.title': '还没有待办',
         'act.empty.lead': '这里会放从会话里整理出的待办、卡住事项和已完成事项。',
         'settings.title': '设置',
-        'settings.subtitle': '本机配置会写入用户配置文件，重启服务后生效。',
+        'settings.subtitle': '本机配置会写入用户配置文件，下次整理时生效。',
         'settings.close': '关闭',
         'settings.language': '界面语言',
         'settings.extractor': '大模型抽取配置',
+        'settings.maxLlmSessions': '每次整理最多调用大模型的会话数',
         'settings.sinceDays': '回溯天数：只抽取最近 N 天内的会话',
         'settings.maxInteractions': '每会话最多交互记录数（一次用户派发→Agent 回复为一条）',
         'settings.apiKeyKeep': '输入新 API key 覆盖，留空保持不变',
@@ -305,7 +315,7 @@
         'settings.apiKeyLabel': 'API key:',
         'settings.save': '保存配置',
         'settings.saving': '保存中...',
-        'settings.savedRestart': '配置已保存，重启后生效。',
+        'settings.savedRestart': '配置已保存，下次整理时生效。',
         'settings.saveFailed': '配置保存失败',
         'act.status.updateFailed': '待办状态更新失败',
         'obs.type.file_read': '读取文件',
diff --git a/src/viewer/parts/app/08-state-expert.js b/src/viewer/parts/app/08-state-expert.js
index 8c9cbefa..3a2ef572 100644
--- a/src/viewer/parts/app/08-state-expert.js
+++ b/src/viewer/parts/app/08-state-expert.js
@@ -72,7 +72,7 @@
       audit: { loaded: false, entries: [], opFilter: '' },
       activity: { loaded: false, observations: [], sessions: [], typeFilter: '', loadingPhase: '', warnings: [] },
       lessons: { loaded: false, items: [], search: '', skillSearch: '', skillRootFilter: 'all', mode: 'explicit', projects: [] },
-      actions: { loaded: false, items: [], reviewItems: [], frontier: [], statusFilter: '', search: '', doneExpanded: false, extractStatus: '', extractMessage: '', extractInFlight: false, stale: false, config: null, configSaving: false, configDraft: {} },
+      actions: { loaded: false, items: [], reviewItems: [], frontier: [], statusFilter: '', search: '', doneExpanded: false, extractStatus: '', extractMessage: '', extractInFlight: false, extractJob: null, stale: false, config: null, configSaving: false, configDraft: {} },
       inbox: { loaded: false, items: [], awaitingItems: [], answeredItems: [], dismissedItems: [], replyingId: null, pendingById: {}, briefingExpanded: false, answeredExpanded: false },
       crystals: { loaded: false, items: [], search: '', lessonMap: {} },
       profile: { loaded: false, projects: [], selectedProject: '', data: null },
@@ -82,4 +82,3 @@
       settings: { open: false },
       ws: null
     };
-
diff --git a/src/viewer/parts/app/60-actions-todo.js b/src/viewer/parts/app/60-actions-todo.js
index a687104f..7c3fbc0f 100644
--- a/src/viewer/parts/app/60-actions-todo.js
+++ b/src/viewer/parts/app/60-actions-todo.js
@@ -30,6 +30,7 @@
       if (state.settings.open) {
         loadTodoExtractorConfig().then(renderSettingsPanel).catch(function() {});
       }
+      apiGet('todo-extract/status').then(syncTodoExtractJob).catch(function() {});
       if (opts.generate === true) startTodoExtraction(opts.force === true);
     }
 
@@ -56,6 +57,43 @@
       return !!result && (result.engine === 'langextract' || result.engine === 'mixed') && !result.llmFallback;
     }
 
+    function todoExtractionErrorMessage(result) {
+      var code = result && (result.errorCode || (result.result && result.result.errorCode));
+      if (code === 'provider_timeout') return t('act.extract.timeout');
+      if (code === 'config_error') return t('act.extract.configError');
+      if (code === 'provider_error' || code === 'llm_unavailable') return t('act.extract.providerError');
+      return t('act.extract.failedExisting');
+    }
+
+    function todoExtractionResultFromJob(job) {
+      if (!job) return null;
+      return job.result || (job.success === true && job.engine ? job : null);
+    }
+
+    function syncTodoExtractJob(job) {
+      if (!job || !job.status || job.status === 'idle') return job;
+      state.actions.extractJob = job;
+      if (job.status === 'running') {
+        state.actions.extractInFlight = true;
+        state.actions.extractStatus = 'running';
+        state.actions.extractMessage = t('act.extract.runningExisting');
+        if (state.activeTab === 'actions') renderActions();
+        return job;
+      }
+      state.actions.extractInFlight = false;
+      if (job.status === 'done') {
+        var result = todoExtractionResultFromJob(job);
+        state.actions.extractStatus = 'done';
+        state.actions.extractFallback = !todoExtractionUsedLlm(result);
+        state.actions.extractMessage = todoExtractionSummary(result);
+      } else if (job.status === 'error') {
+        state.actions.extractStatus = 'error';
+        state.actions.extractMessage = todoExtractionErrorMessage(job);
+      }
+      if (state.activeTab === 'actions') renderActions();
+      return job;
+    }
+
     function refreshActionListsAfterExtract() {
       return Promise.all([
         apiGet('actions'),
@@ -127,6 +165,7 @@
       html += '<input id="todo-config-LANGEXTRACT_BASE_URL" class="search-input" value="' + value('LANGEXTRACT_BASE_URL', 'https://api.novita.ai/openai/v1') + '" placeholder="https://api.novita.ai/openai/v1" />';
       html += '<input id="todo-config-LANGEXTRACT_THINKING_DEPTH" class="search-input" value="' + value('LANGEXTRACT_THINKING_DEPTH', 'medium') + '" placeholder="medium" />';
       html += '<input id="todo-config-AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS" class="search-input" value="' + value('AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS', '120000') + '" placeholder="120000" />';
+      html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.maxLlmSessions')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS', '12') + '" placeholder="12" /></div>';
       html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.sinceDays')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS', '7') + '" placeholder="7" /></div>';
       html += '<div><div class="action-meta-text" style="margin-bottom:4px;">' + esc(t('settings.maxInteractions')) + '</div><input id="todo-config-AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION" class="search-input" type="number" min="1" value="' + value('AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION', '10') + '" placeholder="10" /></div>';
       html += '<div><input id="todo-config-LANGEXTRACT_API_KEY" class="search-input" type="password" placeholder="' + esc(keyLabel) + '" />';
@@ -151,6 +190,7 @@
         'LANGEXTRACT_BASE_URL',
         'LANGEXTRACT_THINKING_DEPTH',
         'AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS',
+        'AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS',
         'AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS',
         'AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION',
         'LANGEXTRACT_API_KEY'
@@ -171,6 +211,7 @@
         'LANGEXTRACT_BASE_URL',
         'LANGEXTRACT_THINKING_DEPTH',
         'AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS',
+        'AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS',
         'AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS',
         'AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION',
         'LANGEXTRACT_API_KEY'
@@ -247,13 +288,21 @@
       // settings would never take effect on this primary extraction path.
       apiPost('todo-extract/generate', {
         force: force === true
-      }).then(function(result) {
+      }).then(function(job) {
+        var result = todoExtractionResultFromJob(job);
+        if (job && job.status === 'running') {
+          state.actions.extractJob = job;
+          state.actions.extractStatus = 'running';
+          state.actions.extractMessage = t('act.extract.runningExisting');
+          return refreshActionListsAfterExtract();
+        }
         var delta = todoExtractionDelta(result);
         if (!result || result.success !== true) {
           state.actions.extractStatus = 'error';
-          state.actions.extractMessage = t('act.extract.failedExisting');
+          state.actions.extractMessage = todoExtractionErrorMessage(job || result);
           return null;
         }
+        state.actions.extractJob = job;
         state.actions.extractStatus = 'done';
         state.actions.extractFallback = !todoExtractionUsedLlm(result);
         state.actions.extractMessage = todoExtractionSummary(result);
@@ -267,7 +316,9 @@
         state.actions.extractMessage = t('act.extract.failedExisting');
       }).then(function() {
         clearTimeout(softRefreshTimer);
-        state.actions.extractInFlight = false;
+        if (!state.actions.extractJob || state.actions.extractJob.status !== 'running') {
+          state.actions.extractInFlight = false;
+        }
         if (state.activeTab === 'actions' && !actionsScrolledAway()) {
           renderActions();
         } else if (state.activeTab !== 'actions') {
diff --git a/src/viewer/server.ts b/src/viewer/server.ts
index 7084046c..c4ba5463 100644
--- a/src/viewer/server.ts
+++ b/src/viewer/server.ts
@@ -12,7 +12,7 @@ import { homedir } from "node:os";
 import { renderViewerDocument } from "./document.js";
 import type { Action, CompressedObservation, Memory, ReviewQueueItem, Session } from "../types.js";
 import { KV, fingerprintId } from "../state/schema.js";
-import { generateTodosFromSessions, updateChangedTodoCards } from "../functions/todo-extract.js";
+import { getTodoExtractJobStatus, runTodoExtractJob, updateChangedTodoCards } from "../functions/todo-extract.js";
 import { getTodoExtractorUserConfig, getUserEnvPath, writeUserEnv, WRITABLE_TODO_EXTRACT_KEYS } from "../config.js";
 
 // Self-host the viewer favicon at /favicon.svg instead of an inline
@@ -1382,11 +1382,16 @@ export function startViewerServer(
     if (method === "POST" && pathname === "/agentmemory/todo-extract/generate") {
       const raw = await readBody(req);
       const body = raw ? JSON.parse(raw) as Record<string, unknown> : {};
-      const result = await generateTodosFromSessions(kv as ViewerKv, body);
+      const result = await runTodoExtractJob(kv as ViewerKv, body);
       json(res, 200, result, req);
       return;
     }
 
+    if (method === "GET" && pathname === "/agentmemory/todo-extract/status") {
+      json(res, 200, getTodoExtractJobStatus(), req);
+      return;
+    }
+
     if (method === "POST" && pathname === "/agentmemory/todo/update") {
       const raw = await readBody(req);
       const body = raw ? JSON.parse(raw) as Record<string, unknown> : {};
@@ -1419,7 +1424,7 @@ export function startViewerServer(
         success: true,
         envPath: getUserEnvPath(),
         config: getTodoExtractorUserConfig(),
-        restartRequired: true,
+        restartRequired: false,
       }, req);
       return;
     }
diff --git a/test/review-action.test.ts b/test/review-action.test.ts
index 296e8a28..66198f7e 100644
--- a/test/review-action.test.ts
+++ b/test/review-action.test.ts
@@ -103,6 +103,7 @@ describe("review action candidates", () => {
     const response = await sdk.trigger("api::todo-extract-generate", req({
       maxSessions: 3,
       maxObservationsPerSession: 20,
+      maxLlmSessions: 4,
       project: "agentmemory-lab",
       force: true,
       cleanup: "dry-run",
@@ -113,6 +114,7 @@ describe("review action candidates", () => {
       success: true,
       maxSessions: 3,
       maxObservationsPerSession: 20,
+      maxLlmSessions: 4,
       project: "agentmemory-lab",
       force: true,
       cleanup: "dry-run",
@@ -124,7 +126,7 @@ describe("review action candidates", () => {
     const oldKey = process.env.LANGEXTRACT_API_KEY;
     process.env.LANGEXTRACT_MODEL = "deepseek/deepseek-v4-flash";
     process.env.LANGEXTRACT_API_KEY = "secret";
-    const response = await sdk.trigger("api::todo-extractor-config", req()) as { status_code: number; body: { success: boolean; config: Record<string, unknown>; envPath: string } };
+    const response = await sdk.trigger("api::todo-extractor-config", req()) as { status_code: number; body: { success: boolean; config: Record<string, unknown>; envPath: string; restartRequired: boolean } };
     if (oldModel === undefined) delete process.env.LANGEXTRACT_MODEL;
     else process.env.LANGEXTRACT_MODEL = oldModel;
     if (oldKey === undefined) delete process.env.LANGEXTRACT_API_KEY;
@@ -136,6 +138,22 @@ describe("review action candidates", () => {
     expect(response.body.config.LANGEXTRACT_MODEL).toBe("deepseek/deepseek-v4-flash");
     expect(response.body.config.LANGEXTRACT_API_KEY).toBeUndefined();
     expect(response.body.config.LANGEXTRACT_API_KEY_CONFIGURED).toBe(true);
+    expect(response.body.restartRequired).toBe(false);
+  });
+
+  it("exposes todo extraction job status through the API", async () => {
+    sdk.registerFunction("mem::todo-extract-status", async () => ({
+      success: true,
+      jobId: "job-1",
+      status: "running",
+      startedAt: "2026-06-24T03:00:00Z",
+      inFlight: true,
+    }));
+
+    const response = await sdk.trigger("api::todo-extract-status", req()) as { status_code: number; body: Record<string, unknown> };
+
+    expect(response.status_code).toBe(200);
+    expect(response.body).toMatchObject({ success: true, jobId: "job-1", status: "running", inFlight: true });
   });
 
   it("rejects invalid todo extraction limits through the API", async () => {
diff --git a/test/session-highlights-api.test.ts b/test/session-highlights-api.test.ts
index 6417a87b..bb11fabc 100644
--- a/test/session-highlights-api.test.ts
+++ b/test/session-highlights-api.test.ts
@@ -30,9 +30,9 @@ describe("session highlights REST API wiring", () => {
     const agents = readText("AGENTS.md");
     const index = readText("src/index.ts");
 
-    expect(endpointCount).toBe(141);
-    expect(readme).toContain("141 endpoints on port");
-    expect(agents).toContain("141 REST endpoints");
-    expect(index).toContain("REST API: 141 endpoints");
+    expect(endpointCount).toBe(142);
+    expect(readme).toContain("142 endpoints on port");
+    expect(agents).toContain("142 REST endpoints");
+    expect(index).toContain("REST API: 142 endpoints");
   });
 });
diff --git a/test/todo-extract.test.ts b/test/todo-extract.test.ts
index 68856745..393d900e 100644
--- a/test/todo-extract.test.ts
+++ b/test/todo-extract.test.ts
@@ -3,9 +3,10 @@ import { describe, expect, it, beforeEach, afterEach, vi } from "vitest";
 vi.mock("../src/config.js", () => ({
   DEFAULT_LANGEXTRACT_BASE_URL: "https://api.novita.ai/openai/v1",
   DEFAULT_TODO_EXTRACT_TIMEOUT_MS: 120_000,
+  DEFAULT_TODO_EXTRACT_MAX_LLM_SESSIONS: 12,
   DEFAULT_TODO_EXTRACT_SINCE_DAYS: 7,
   DEFAULT_TODO_EXTRACT_MAX_INTERACTIONS: 10,
-  DEFAULT_TODO_EXTRACT_MAX_SESSIONS: 8,
+  DEFAULT_TODO_EXTRACT_MAX_SESSIONS: Number.POSITIVE_INFINITY,
   getEnvVar: (key: string) => {
     const values: Record<string, string> = {
       AGENTMEMORY_TODO_EXTRACTOR: "rules",
@@ -21,7 +22,7 @@ vi.mock("../src/config.js", () => ({
   normalizeTodoExtractorProvider: (value?: string) => (value || "openai").toLowerCase(),
 }));
 
-import { cleanPollutedTodoCards, updateChangedTodoCards, cleanTodoTitle, generateTodosFromSessions, validateTodoEvidence, runLangExtractSidecar, type ExtractedTodo } from "../src/functions/todo-extract.js";
+import { cleanPollutedTodoCards, updateChangedTodoCards, cleanTodoTitle, generateTodosFromSessions, getTodoExtractJobStatus, runTodoExtractJob, startTodoExtractJob, validateTodoEvidence, runLangExtractSidecar, type ExtractedTodo } from "../src/functions/todo-extract.js";
 import type { Action, CompressedObservation, ReviewQueueItem, Session } from "../src/types.js";
 import { KV } from "../src/state/schema.js";
 import { mockKV } from "./helpers/mocks.js";
@@ -864,6 +865,9 @@ describe("todo extraction scope — sinceDays + interaction window (STEP-11)", (
   afterEach(() => {
     delete process.env.AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS;
     delete process.env.AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION;
+    delete process.env.AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS;
+    delete process.env.AGENTMEMORY_TODO_EXTRACTOR;
+    delete process.env.LANGEXTRACT_PYTHON;
   });
 
   // Two completed sessions, both inside the 14d "recent" bucket (so neither is
@@ -912,6 +916,38 @@ describe("todo extraction scope — sinceDays + interaction window (STEP-11)", (
     expect((await kv.list<Action>(KV.actions))[0].metadata?.todoExtraction).toMatchObject({ sourceSessionId: "ses_recent" });
   });
 
+  it("does not cap eligible sessions by default after applying the sinceDays window", async () => {
+    const narratives = [
+      "后续需要修复登录接口的超时问题。",
+      "后续需要更新数据库驱动到 v5。",
+      "后续需要补充新用户的上手文档。",
+      "后续需要确认头像资源加载。",
+      "后续需要修复设置面板保存问题。",
+      "后续需要补充会话列表分页。",
+      "后续需要校验证据引用。",
+      "后续需要调整更新按钮文案。",
+      "后续需要补充端口占用提示。",
+    ];
+    for (let i = 0; i < narratives.length; i++) {
+      const id = `ses_window_${i + 1}`;
+      const at = daysAgo(1 + i / 100);
+      await kv.set(KV.sessions, id, session({
+        id, status: "completed", startedAt: at, endedAt: at, observationCount: 1,
+      }));
+      await kv.set(KV.observations(id), `o_window_${i + 1}`, obs({
+        id: `o_window_${i + 1}`, sessionId: id, timestamp: at, narrative: narratives[i],
+      }));
+    }
+
+    const result = await generateTodosFromSessions(kv as never, {
+      force: true, scanSources: false, cleanup: "none", sinceDays: 7,
+    });
+
+    expect(result.scannedSessions).toBe(9);
+    const checkpoint = await kv.get<{ cursor: string }>(KV.scanCheckpoints, "todo-extract:all");
+    expect(Object.keys(JSON.parse(checkpoint?.cursor || "{}"))).toHaveLength(9);
+  });
+
   // A user message in synthetic-compression form: type "conversation", title
   // === raw hookType "prompt_submit" (the interaction-boundary signal).
   function promptObs(id: string, narrative: string, timestamp: string): CompressedObservation {
@@ -940,6 +976,100 @@ describe("todo extraction scope — sinceDays + interaction window (STEP-11)", (
     expect((await kv.list<Action>(KV.actions))[0].sourceObservationIds).toEqual(["t3"]);
   });
 
+  it("defaults to the most recent 10 interaction records per session", async () => {
+    process.env.AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS = "30";
+    await kv.set(KV.sessions, "ses_turns", session({ id: "ses_turns", status: "active", observationCount: 12 }));
+    const narratives = [
+      "后续需要修复登录接口超时。",
+      "后续需要更新数据库驱动到 v5。",
+      "后续需要创建导出报告入口。",
+      "后续需要移除废弃配置开关。",
+      "后续需要实现离线缓存同步。",
+      "后续需要补充安装故障文档。",
+      "后续需要处理浏览器插件认证失败。",
+      "后续需要调整设置保存流程。",
+      "后续需要验证头像 PNG 加载。",
+      "后续需要重试 CI 发布流程。",
+      "后续需要排查工作台空白页。",
+      "后续需要整理权限错误提示。",
+    ];
+    for (let i = 0; i < narratives.length; i++) {
+      await kv.set(KV.observations("ses_turns"), `t${i + 1}`, promptObs(`t${i + 1}`, narratives[i], daysAgo(12 - i)));
+    }
+
+    const result = await generateTodosFromSessions(kv as never, { force: true, scanSources: false, cleanup: "none" });
+
+    expect(result.scannedObservations).toBe(10);
+    const sourceIds = (await kv.list<Action>(KV.actions)).flatMap((action) => action.sourceObservationIds || []);
+    expect(sourceIds).not.toContain("t1");
+    expect(sourceIds).not.toContain("t2");
+    const checkpoint = await kv.get<{ cursor: string }>(KV.scanCheckpoints, "todo-extract:all");
+    expect(JSON.parse(checkpoint?.cursor || "{}")).toHaveProperty("ses_turns");
+  });
+
+  it("bounds LLM sidecar calls while still scanning the full session window", async () => {
+    process.env.AGENTMEMORY_TODO_EXTRACTOR = "langextract";
+    process.env.AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS = "1";
+    process.env.LANGEXTRACT_PYTHON = "definitely-missing-python";
+    for (let i = 0; i < 3; i++) {
+      const id = `ses_llm_budget_${i + 1}`;
+      const at = daysAgo(i + 1);
+      await kv.set(KV.sessions, id, session({ id, status: "active", startedAt: at, endedAt: at, observationCount: 1 }));
+      await kv.set(KV.observations(id), `o_llm_budget_${i + 1}`, obs({
+        id: `o_llm_budget_${i + 1}`,
+        sessionId: id,
+        timestamp: at,
+        narrative: `后续需要修复第 ${i + 1} 个抽取预算测试问题。`,
+      }));
+    }
+
+    const result = await generateTodosFromSessions(kv as never, {
+      force: true, scanSources: false, cleanup: "none", sinceDays: 30,
+    });
+
+    expect(result.scannedSessions).toBe(3);
+    expect(result.processedSessions).toBe(3);
+    expect(result.llmSessionBudget).toBe(1);
+    expect(result.llmSessionsAttempted).toBe(1);
+    expect(result.llmSessionsSkipped).toBe(2);
+    expect(result.llmFallback).toBe(true);
+    expect(result.errorCode).toBeDefined();
+    delete process.env.AGENTMEMORY_TODO_EXTRACTOR;
+    delete process.env.AGENTMEMORY_TODO_EXTRACT_MAX_LLM_SESSIONS;
+    delete process.env.LANGEXTRACT_PYTHON;
+  });
+
+  it("exposes a single running extraction job instead of starting duplicate work", async () => {
+    await kv.set(KV.sessions, "ses_job", session({ id: "ses_job", status: "active", observationCount: 1 }));
+    await kv.set(KV.observations("ses_job"), "o_job", obs({
+      id: "o_job", sessionId: "ses_job", narrative: "后续需要验证抽取任务单飞。",
+    }));
+
+    const first = await startTodoExtractJob(kv as never, { force: true, scanSources: false, cleanup: "none" });
+    const second = await startTodoExtractJob(kv as never, { force: true, scanSources: false, cleanup: "none" });
+
+    expect(first.status).toBe("running");
+    expect(second.status).toBe("running");
+    expect(second.inFlight).toBe(true);
+    expect(second.jobId).toBe(first.jobId);
+    await vi.waitFor(() => expect(getTodoExtractJobStatus().status).toBe("done"));
+  });
+
+  it("waits for the owner extraction request but lets duplicate requests observe the running job", async () => {
+    await kv.set(KV.sessions, "ses_run_job", session({ id: "ses_run_job", status: "active", observationCount: 1 }));
+    await kv.set(KV.observations("ses_run_job"), "o_run_job", obs({
+      id: "o_run_job", sessionId: "ses_run_job", narrative: "后续需要验证抽取请求等待最终结果。",
+    }));
+
+    const owner = runTodoExtractJob(kv as never, { force: true, scanSources: false, cleanup: "none" });
+    const duplicate = await runTodoExtractJob(kv as never, { force: true, scanSources: false, cleanup: "none" });
+    const completed = await owner;
+
+    expect(duplicate).toMatchObject({ status: "running", inFlight: true });
+    expect(completed.status).toBe("done");
+    expect(completed.result).toMatchObject({ success: true, processedSessions: 1 });
+  });
+
   it("treats a session with no user-message boundary as a single interaction", async () => {
     process.env.AGENTMEMORY_TODO_EXTRACT_SINCE_DAYS = "30";
     process.env.AGENTMEMORY_TODO_EXTRACT_MAX_INTERACTIONS_PER_SESSION = "1";
diff --git a/test/viewer-session-id.test.ts b/test/viewer-session-id.test.ts
index 45da7b22..e499c5ae 100644
--- a/test/viewer-session-id.test.ts
+++ b/test/viewer-session-id.test.ts
@@ -1075,10 +1075,10 @@ describe("viewer session rendering", () => {
     target.getAttribute = (name: string) => name === "data-action" ? "save-todo-config" : null;
     target.closest = (selector: string) => selector === "[data-action]" ? target : null;
     dispatchDocumentClick(target);
-    await waitFor(() => sandbox.state.actions.extractMessage === "Config saved. Restart the service to apply it.");
+    await waitFor(() => sandbox.state.actions.extractMessage === "Config saved. It applies to the next organize run.");
 
     expect(posts[0]).toMatchObject({ LANGEXTRACT_MODEL: "deepseek/deepseek-v4-flash", AGENTMEMORY_TODO_EXTRACT_TIMEOUT_MS: "120000", LANGEXTRACT_API_KEY: "secret" });
-    expect(sandbox.state.actions.extractMessage).toBe("Config saved. Restart the service to apply it.");
+    expect(sandbox.state.actions.extractMessage).toBe("Config saved. It applies to the next organize run.");
   });
 
   it("keeps unsaved todo extractor config while the settings panel rerenders", () => {
@@ -1212,6 +1212,29 @@ describe("viewer session rendering", () => {
     expect(sandbox.state.actions.items[0].title).toBe("整理首版功能文档");
   });
 
+  it("restores running todo extraction state after the actions view reloads", async () => {
+    const { sandbox } = loadViewerSandbox();
+    sandbox.fetch = async (input: unknown) => {
+      const url = String(input);
+      if (url.includes("todo-extract/status")) {
+        return { ok: true, json: async () => ({ success: true, jobId: "job-1", status: "running", startedAt: "2026-06-24T03:00:00Z", inFlight: true }) };
+      }
+      if (url.includes("review?status=pending")) return { ok: true, json: async () => ({ items: [] }) };
+      if (url.includes("frontier")) return { ok: true, json: async () => ({ frontier: [] }) };
+      if (url.includes("actions")) return { ok: true, json: async () => ({ actions: [] }) };
+      if (url.includes("inbox")) return { ok: true, json: async () => ({ items: [] }) };
+      return { ok: true, json: async () => ({}) };
+    };
+
+    sandbox.state.activeTab = "actions";
+    await sandbox.loadActions();
+    await waitFor(() => sandbox.state.actions.extractInFlight === true);
+
+    expect(sandbox.state.actions.extractInFlight).toBe(true);
+    expect(sandbox.state.actions.extractStatus).toBe("running");
+    expect(sandbox.state.actions.extractMessage).toBe("Still organizing from a previous request...");
+  });
+
   it("renders the action classification metrics without a false waiting section when inbox is empty", () => {
     const { sandbox, getElement } = loadViewerSandbox();
     sandbox.state.activeTab = "actions";