diff --git a/apps/desktop/DESIGN.md b/apps/desktop/DESIGN.md
index 5084f20e74..24eea03278 100644
--- a/apps/desktop/DESIGN.md
+++ b/apps/desktop/DESIGN.md
@@ -155,6 +155,8 @@ fill/shadow), `ghost`, `floating` (a control loose from any surface — opaque
 popover fill + `shadow-md`, hover lifts the glyph only), `link`, `text`
 (boxless quiet inline — "Cancel", "Clear"), `textStrong` (bold underlined
 inline affordance — "Change", "Open logs").
+`grip` is the quiet, fill-free drawer handle; pair it with size `grip` for a
+48×16 hit area around a small horizontal ridge.
 
 **Sizes:** `default`, `xs`, `sm`, `lg`, `inline` (flush, zero box — for buttons
 that sit inside a heading/sentence; replaces `h-auto px-0 py-0`), `micro`
@@ -347,6 +349,10 @@ so glass and message-bubble transparency do not reveal scrolling text.
   pause/resume preserve the user's disclosure choice. Error banners meet the
   stack's top edge without a blank padding strip. File and preview links remain
   visible at the bottom of the stack, below the queue and all status groups.
+  A centered ridge on the composer's top edge hides/reveals the entire stack,
+  including the git row, with a short downward/upward drawer slide. Its choice
+  persists per conversation and owner, not globally. Hidden sections stay
+  mounted but inert so their disclosure choices survive; reduced motion is instant.
 - Popping out a composer makes it the window's only visible composer. It keeps
   its viewport placement while hover or keyboard focus selects a chat pane;
   moving back into the editor retains that recipient. Drafts, attachments and
diff --git a/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.test.tsx b/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.test.tsx
new file mode 100644
index 0000000000..f414f7af50
--- /dev/null
+++ b/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.test.tsx
@@ -0,0 +1,31 @@
+import { act, renderHook } from '@testing-library/react'
+import { beforeEach, expect, it, vi } from 'vitest'
+
+import { $collapsedStatusDrawers } from '@/store/composer-status-drawer'
+
+import { useStatusDrawer } from './use-status-drawer'
+
+vi.mock('@/store/session-states', () => ({
+  knownOwnerForSession: () => ({ connectionId: 'local', profile: 'default', targetProfile: 'default' })
+}))
+
+beforeEach(() => $collapsedStatusDrawers.set([]))
+
+it('keeps stored conversations independent from drafts and other conversations', () => {
+  const { result, rerender } = renderHook(({ sessionKey }) => useStatusDrawer(sessionKey), {
+    initialProps: { sessionKey: null as string | null }
+  })
+
+  act(() => result.current.toggle())
+  expect(result.current.collapsed).toBe(true)
+
+  rerender({ sessionKey: 'existing-a' })
+  expect(result.current.collapsed).toBe(false)
+  act(() => result.current.toggle())
+  rerender({ sessionKey: 'existing-b' })
+  expect(result.current.collapsed).toBe(false)
+  rerender({ sessionKey: 'existing-a' })
+  expect(result.current.collapsed).toBe(true)
+  rerender({ sessionKey: null })
+  expect(result.current.collapsed).toBe(false)
+})
diff --git a/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.ts b/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.ts
new file mode 100644
index 0000000000..27985d6221
--- /dev/null
+++ b/apps/desktop/src/app/chat/composer/hooks/use-status-drawer.ts
@@ -0,0 +1,50 @@
+import { useStore } from '@nanostores/react'
+import { useState } from 'react'
+
+import { useStoreSelector } from '@/lib/use-session-slice'
+import { $collapsedStatusDrawers, setStatusDrawerCollapsed, statusDrawerKey } from '@/store/composer-status-drawer'
+import { $activeConnectionId } from '@/store/connections'
+import { $activeGatewayProfile, normalizeProfileKey } from '@/store/profile'
+import { isSessionOwnerRoute } from '@/store/session-request-router'
+import { knownOwnerForSession } from '@/store/session-states'
+
+import { useComposerScope } from '../scope'
+
+export function useStatusDrawer(sessionKey: string | null) {
+  const scope = useComposerScope()
+  const activeProfile = useStore($activeGatewayProfile)
+  const activeConnection = useStore($activeConnectionId)
+  const owner = knownOwnerForSession(sessionKey)
+  const route = isSessionOwnerRoute(owner) ? owner : null
+
+  const profile = normalizeProfileKey(
+    route?.profile ?? (typeof owner === 'string' ? owner : scope.profile ?? activeProfile)
+  )
+
+  const targetProfile = normalizeProfileKey(route?.targetProfile ?? profile)
+  const connectionId = route ? route.connectionId ?? null : owner ? null : scope.connectionId ?? activeConnection
+  const ownerKey = JSON.stringify([connectionId, profile, targetProfile])
+  const key = sessionKey ? statusDrawerKey({ connectionId, profile, targetProfile, sessionId: sessionKey }) : null
+  const saved = useStoreSelector($collapsedStatusDrawers, keys => key !== null && keys.includes(key))
+  const draftKey = key ?? ownerKey
+  const [draft, setDraft] = useState({ key: draftKey, collapsed: false })
+
+  // Unsaved drafts have no durable preference. Never carry their choice into
+  // a stored conversation selected from the sidebar or a subsequent new chat.
+  if (draft.key !== draftKey) {
+    setDraft({ key: draftKey, collapsed: false })
+  }
+
+  const collapsed = key ? saved : draft.key === draftKey && draft.collapsed
+
+  return {
+    collapsed,
+    toggle: () => {
+      if (key) {
+        setStatusDrawerCollapsed(key, !collapsed)
+      } else {
+        setDraft({ key: draftKey, collapsed: !collapsed })
+      }
+    }
+  }
+}
diff --git a/apps/desktop/src/app/chat/composer/index.tsx b/apps/desktop/src/app/chat/composer/index.tsx
index 249f8e5f72..9d2a953f1d 100644
--- a/apps/desktop/src/app/chat/composer/index.tsx
+++ b/apps/desktop/src/app/chat/composer/index.tsx
@@ -1,6 +1,15 @@
 import { ComposerPrimitive } from '@assistant-ui/react'
 import { useStore } from '@nanostores/react'
-import { type ClipboardEvent, type FormEvent, type KeyboardEvent, useCallback, useEffect, useMemo, useRef } from 'react'
+import {
+  type ClipboardEvent,
+  type FormEvent,
+  type KeyboardEvent,
+  useCallback,
+  useEffect,
+  useId,
+  useMemo,
+  useRef
+} from 'react'
 
 import { useTourMarker } from '@/app/chat/tour-marker'
 import { useHudComposerDrag } from '@/app/hud/composer-drag'
@@ -66,6 +75,7 @@ import { useComposerVoice } from './hooks/use-composer-voice'
 import { useEmojiCompletions } from './hooks/use-emoji-completions'
 import { useComposerMicroActions } from './hooks/use-micro-actions'
 import { useSlashCompletions } from './hooks/use-slash-completions'
+import { useStatusDrawer } from './hooks/use-status-drawer'
 import { useSessionStatusPresence } from './hooks/use-status-presence'
 import { shouldConvertPasteToAttachment } from './large-paste'
 import { ActionBadges } from './micro-actions'
@@ -84,6 +94,7 @@ import {
 import { useComposerScope, useComposerSurfaceId } from './scope'
 import { ComposerStatusStack } from './status-stack'
 import { CodingStatusRow } from './status-stack/coding-row'
+import { StatusDrawerContent, StatusDrawerToggle } from './status-stack/drawer'
 import { SuggestionPills } from './suggestion-pills'
 import { extractClipboardImageBlobs, openDirectiveScope } from './text-utils'
 import { ComposerTriggerPopover } from './trigger-popover'
@@ -193,6 +204,9 @@ export function ChatBar({
   // undelivered behind the blocked tool batch). Drives the button affordance.
   const blockingPrompt = useStore(useMemo(() => sessionBlockingPrompt(sessionId ?? null), [sessionId]))
   const activeQueueSessionKey = queueSessionKey || sessionId || null
+  const { collapsed: statusDrawerCollapsed, toggle: toggleStatusDrawer } = useStatusDrawer(activeQueueSessionKey)
+  const statusDrawerId = useId()
+  const codingDrawerId = useId()
 
   // Status items (subagents, background processes) are keyed by the RUNTIME
   // session id — gateway events and process.list both speak that id. Only the
@@ -1298,37 +1312,39 @@ export function ChatBar({
               queue). An in-flow dock child: the dock is bottom-anchored, so it
               grows upward over the thread and the dock's own measurement covers
               it. Collapses to nothing when every status is empty. */}
-          <ComposerStatusStack
-            onSubmit={onSubmit}
-            queue={
-              activeQueueSessionKey && queuedPrompts.length > 0 ? (
-                <QueuePanel
-                  busy={busy}
-                  editingId={queueEdit?.entryId ?? null}
-                  entries={queuedPrompts}
-                  onDelete={id => {
-                    if (removeQueuedPrompt(activeQueueSessionKey, id) && queueEdit?.entryId === id) {
-                      exitQueuedEdit('cancel')
-                    }
-                  }}
-                  onEdit={beginQueuedEdit}
-                  onResume={() => {
-                    unparkQueuedPrompts(activeQueueSessionKey)
+          <StatusDrawerContent collapsed={statusDrawerCollapsed} id={statusDrawerId}>
+            <ComposerStatusStack
+              onSubmit={onSubmit}
+              queue={
+                activeQueueSessionKey && queuedPrompts.length > 0 ? (
+                  <QueuePanel
+                    busy={busy}
+                    editingId={queueEdit?.entryId ?? null}
+                    entries={queuedPrompts}
+                    onDelete={id => {
+                      if (removeQueuedPrompt(activeQueueSessionKey, id) && queueEdit?.entryId === id) {
+                        exitQueuedEdit('cancel')
+                      }
+                    }}
+                    onEdit={beginQueuedEdit}
+                    onResume={() => {
+                      unparkQueuedPrompts(activeQueueSessionKey)
 
-                    // Idle → kick the head immediately; busy → the settle drain
-                    // takes over now that the park is lifted.
-                    if (!busy) {
-                      void drainNextQueued()
-                    }
-                  }}
-                  onSendNow={id => void sendQueuedNow(id)}
-                  onSteerNow={id => void steerQueuedNow(id)}
-                  parked={queueParked}
-                />
-              ) : null
-            }
-            sessionId={statusSessionId}
-          />
+                      // Idle → kick the head immediately; busy → the settle drain
+                      // takes over now that the park is lifted.
+                      if (!busy) {
+                        void drainNextQueued()
+                      }
+                    }}
+                    onSendNow={id => void sendQueuedNow(id)}
+                    onSteerNow={id => void steerQueuedNow(id)}
+                    parked={queueParked}
+                  />
+                ) : null
+              }
+              sessionId={statusSessionId}
+            />
+          </StatusDrawerContent>
           <ComposerPrimitive.Root
             className={cn(
               'group/composer relative w-full overflow-visible rounded-2xl',
@@ -1344,7 +1360,7 @@ export function ChatBar({
             data-hud-grabbing={hudGrabbing ? '' : undefined}
             data-popped-out={poppedOut ? '' : undefined}
             data-slot="composer-root"
-            data-status-stack={statusStackVisible ? '' : undefined}
+            data-status-stack={statusStackVisible && !statusDrawerCollapsed ? '' : undefined}
             data-thread-scrolled-up={scrolledUp ? '' : undefined}
             data-tip-region=""
             data-tour={composerTourMarker}
@@ -1398,6 +1414,13 @@ export function ChatBar({
               />
             )}
             <div className="relative w-full rounded-[inherit]">
+              {!hudMode && !guidedChat && (
+                <StatusDrawerToggle
+                  collapsed={statusDrawerCollapsed}
+                  controls={`${statusDrawerId} ${codingDrawerId}`}
+                  onToggle={toggleStatusDrawer}
+                />
+              )}
               {hudMode && busy && <span aria-hidden className="arc-border arc-composer" />}
               <div
                 className={cn(
@@ -1416,21 +1439,23 @@ export function ChatBar({
               >
                 <div aria-hidden className={composerInputBacking} />
                 {!guidedChat && (
-                  <CodingStatusRow
-                    onBranchOff={handleBranchOff}
-                    onConvertBranch={handleConvertBranch}
-                    onListBranches={handleListBranches}
-                    // A tile's rail reviews ITS worktree: pin the pane's scope to
-                    // this surface's cwd. Main keeps the classic follow-the-
-                    // active-session scope (null).
-                    onOpen={() => toggleReview(scope.target === 'main' ? null : (cwd ?? null), scope.target)}
-                    onOpenWorktree={openInWorktree}
-                    onSwitchBranch={handleSwitchBranch}
-                    // Blank in a bot chat: the row hides itself without a repo,
-                    // and stops probing git / GitHub for a surface that has no
-                    // branch to show. Cheaper than a second composer.
-                    repoPath={botChat ? undefined : cwd}
-                  />
+                  <StatusDrawerContent collapsed={statusDrawerCollapsed} id={codingDrawerId}>
+                    <CodingStatusRow
+                      onBranchOff={handleBranchOff}
+                      onConvertBranch={handleConvertBranch}
+                      onListBranches={handleListBranches}
+                      // A tile's rail reviews ITS worktree: pin the pane's scope to
+                      // this surface's cwd. Main keeps the classic follow-the-
+                      // active-session scope (null).
+                      onOpen={() => toggleReview(scope.target === 'main' ? null : (cwd ?? null), scope.target)}
+                      onOpenWorktree={openInWorktree}
+                      onSwitchBranch={handleSwitchBranch}
+                      // Blank in a bot chat: the row hides itself without a repo,
+                      // and stops probing git / GitHub for a surface that has no
+                      // branch to show. Cheaper than a second composer.
+                      repoPath={botChat ? undefined : cwd}
+                    />
+                  </StatusDrawerContent>
                 )}
                 <div
                   className={cn(
diff --git a/apps/desktop/src/app/chat/composer/status-stack/drawer.css b/apps/desktop/src/app/chat/composer/status-stack/drawer.css
new file mode 100644
index 0000000000..dc81dbd962
--- /dev/null
+++ b/apps/desktop/src/app/chat/composer/status-stack/drawer.css
@@ -0,0 +1,51 @@
+.status-drawer {
+  display: grid;
+  grid-template-rows: 1fr;
+  /* Quick lift, then a soft landing. Closing is deliberately shorter. */
+  transition: grid-template-rows 260ms cubic-bezier(0.16, 1, 0.3, 1);
+}
+
+.status-drawer-clip {
+  min-height: 0;
+  overflow: hidden;
+}
+
+.status-drawer-content {
+  /* The bottom-anchored dock already slides this content as the row opens.
+     Another translate doubles its travel and hides the first half of arrival. */
+  visibility: visible;
+  transition: visibility 0s;
+}
+
+.status-drawer[data-collapsed] {
+  grid-template-rows: 0fr;
+  transition: grid-template-rows 150ms cubic-bezier(0.4, 0, 0.2, 1);
+}
+
+.status-drawer[data-collapsed] .status-drawer-content {
+  visibility: hidden;
+  transition: visibility 0s 150ms;
+}
+
+.status-drawer-toggle {
+  opacity: 0.35;
+  transition: opacity 120ms ease-out, color 120ms ease-out;
+}
+
+.status-drawer-toggle:is(:hover, :focus-visible) {
+  opacity: 1;
+}
+
+/* Hidden content stays mounted, keeping the grip available to reopen it. */
+[data-slot='composer-dock']:not(:has([data-slot='composer-status-stack'], .coding-status-bar)) .status-drawer-toggle {
+  display: none;
+}
+
+@media (prefers-reduced-motion: reduce) {
+  .status-drawer,
+  .status-drawer[data-collapsed],
+  .status-drawer-content,
+  .status-drawer[data-collapsed] .status-drawer-content {
+    transition: none;
+  }
+}
diff --git a/apps/desktop/src/app/chat/composer/status-stack/drawer.tsx b/apps/desktop/src/app/chat/composer/status-stack/drawer.tsx
new file mode 100644
index 0000000000..8fd9807f3b
--- /dev/null
+++ b/apps/desktop/src/app/chat/composer/status-stack/drawer.tsx
@@ -0,0 +1,59 @@
+import './drawer.css'
+
+import type { ReactNode } from 'react'
+
+import { Button } from '@/components/ui/button'
+import { Tip } from '@/components/ui/tooltip'
+import { useI18n } from '@/i18n'
+
+interface StatusDrawerContentProps {
+  children: ReactNode
+  collapsed: boolean
+  id: string
+}
+
+/** Keep the live sections mounted: hiding the drawer must not reset their disclosures. */
+export function StatusDrawerContent({ children, collapsed, id }: StatusDrawerContentProps) {
+  return (
+    <div
+      aria-hidden={collapsed || undefined}
+      className="status-drawer"
+      data-collapsed={collapsed ? '' : undefined}
+      id={id}
+      inert={collapsed}
+    >
+      <div className="status-drawer-clip">
+        <div className="status-drawer-content">{children}</div>
+      </div>
+    </div>
+  )
+}
+
+interface StatusDrawerToggleProps {
+  collapsed: boolean
+  controls: string
+  onToggle: () => void
+}
+
+export function StatusDrawerToggle({ collapsed, controls, onToggle }: StatusDrawerToggleProps) {
+  const { t } = useI18n()
+  const label = collapsed ? t.statusStack.showStack : t.statusStack.hideStack
+
+  return (
+    <Tip label={label}>
+      <Button
+        aria-controls={controls}
+        aria-expanded={!collapsed}
+        aria-label={label}
+        className="status-drawer-toggle absolute -top-2 left-1/2 z-10 -translate-x-1/2"
+        data-slot="status-drawer-toggle"
+        onClick={onToggle}
+        size="grip"
+        type="button"
+        variant="grip"
+      >
+        <span aria-hidden className="h-0.75 w-7 rounded-full bg-current" />
+      </Button>
+    </Tip>
+  )
+}
diff --git a/apps/desktop/src/components/ui/button.tsx b/apps/desktop/src/components/ui/button.tsx
index fdbb64b8c0..97fe05a740 100644
--- a/apps/desktop/src/components/ui/button.tsx
+++ b/apps/desktop/src/components/ui/button.tsx
@@ -26,6 +26,7 @@ const buttonVariants = cva(
         secondary:
           'bg-(--ui-bg-quaternary) text-(--ui-text-primary) hover:bg-(--chrome-action-hover) hover:text-(--ui-text-primary)',
         ghost: 'text-(--ui-text-secondary) hover:bg-(--chrome-action-hover) hover:text-(--ui-text-primary)',
+        grip: 'bg-transparent text-(--ui-text-tertiary) transition-colors hover:text-(--ui-text-secondary) focus-visible:text-(--ui-text-primary)',
         // A control floating free of any surface (fan-menu discs, detached
         // chips): the menu/popover treatment — opaque popover fill + the
         // shared `shadow-md` ring-and-drop. Hover only lifts the glyph; a fill
@@ -55,6 +56,7 @@ const buttonVariants = cva(
         'icon-xs': "size-6 rounded-[4px] [&_svg:not([class*='size-'])]:size-3",
         'icon-sm': 'size-8 rounded-[4px]',
         'icon-lg': 'size-10 rounded-[4px]',
+        grip: 'h-4 w-12 rounded-full p-0',
         'icon-titlebar':
           'titlebar-icon-button h-(--titlebar-control-height) w-(--titlebar-control-size) rounded-[4px] [&_svg:not([class*="size-"])]:size-(--titlebar-icon-size)'
       }
diff --git a/apps/desktop/src/i18n/ar.ts b/apps/desktop/src/i18n/ar.ts
index 82a14780a4..ae205771d0 100644
--- a/apps/desktop/src/i18n/ar.ts
+++ b/apps/desktop/src/i18n/ar.ts
@@ -2206,6 +2206,8 @@ export const ar = defineLocale({
     }
   },
   statusStack: {
+    hideStack: 'إخفاء لوحة الحالة',
+    showStack: 'إظهار لوحة الحالة',
     agents: 'الوكلاء',
     background: count => `${count} في الخلفية`,
     goalActive: 'الهدف نشط',
diff --git a/apps/desktop/src/i18n/en.ts b/apps/desktop/src/i18n/en.ts
index f173b8d65d..61b337d54c 100644
--- a/apps/desktop/src/i18n/en.ts
+++ b/apps/desktop/src/i18n/en.ts
@@ -3158,6 +3158,8 @@ export const en: Translations = {
   },
 
   statusStack: {
+    hideStack: 'Hide status stack',
+    showStack: 'Show status stack',
     agents: 'Agents',
     background: count => `${count} Background`,
     goalActive: 'Goal active',
diff --git a/apps/desktop/src/i18n/ja.ts b/apps/desktop/src/i18n/ja.ts
index 2357178ac5..f062c3b744 100644
--- a/apps/desktop/src/i18n/ja.ts
+++ b/apps/desktop/src/i18n/ja.ts
@@ -2584,6 +2584,8 @@ export const ja = defineLocale({
   },
 
   statusStack: {
+    hideStack: 'ステータススタックを隠す',
+    showStack: 'ステータススタックを表示',
     agents: 'エージェント',
     background: count => `バックグラウンド ${count} 件`,
     goalActive: '目標進行中',
diff --git a/apps/desktop/src/i18n/ru.ts b/apps/desktop/src/i18n/ru.ts
index 04c5521f72..a7ec6f02ab 100644
--- a/apps/desktop/src/i18n/ru.ts
+++ b/apps/desktop/src/i18n/ru.ts
@@ -2889,6 +2889,8 @@ export const ru = defineLocale({
     }
   },
   statusStack: {
+    hideStack: 'Скрыть панель состояния',
+    showStack: 'Показать панель состояния',
     agents: 'Агенты',
     background: count => `${count} ${RU_NOUN(count, 'фоновая задача', 'фоновые задачи', 'фоновых задач')}`,
     goalActive: 'Цель активна',
diff --git a/apps/desktop/src/i18n/types.ts b/apps/desktop/src/i18n/types.ts
index 158974cbce..2bee181c1e 100644
--- a/apps/desktop/src/i18n/types.ts
+++ b/apps/desktop/src/i18n/types.ts
@@ -2662,6 +2662,8 @@ export interface Translations {
   }
 
   statusStack: {
+    hideStack: string
+    showStack: string
     agents: string
     background: (count: number) => string
     goalActive: string
diff --git a/apps/desktop/src/i18n/zh-hant.ts b/apps/desktop/src/i18n/zh-hant.ts
index 11315e3d59..532349f027 100644
--- a/apps/desktop/src/i18n/zh-hant.ts
+++ b/apps/desktop/src/i18n/zh-hant.ts
@@ -2569,6 +2569,8 @@ export const zhHant = defineLocale({
   },
 
   statusStack: {
+    hideStack: '隱藏狀態面板',
+    showStack: '顯示狀態面板',
     agents: '代理',
     background: count => `${count} 個背景任務`,
     goalActive: '目標進行中',
diff --git a/apps/desktop/src/i18n/zh.ts b/apps/desktop/src/i18n/zh.ts
index 27d0a6c7a9..e10a62d137 100644
--- a/apps/desktop/src/i18n/zh.ts
+++ b/apps/desktop/src/i18n/zh.ts
@@ -3299,6 +3299,8 @@ export const zh = defineLocale({
   },
 
   statusStack: {
+    hideStack: '隐藏状态面板',
+    showStack: '显示状态面板',
     agents: '代理',
     background: count => `${count} 个后台任务`,
     goalActive: '目标进行中',
diff --git a/apps/desktop/src/lib/keybinds/composer-focus-keys.test.ts b/apps/desktop/src/lib/keybinds/composer-focus-keys.test.ts
index c1fe51484c..98ed53b158 100644
--- a/apps/desktop/src/lib/keybinds/composer-focus-keys.test.ts
+++ b/apps/desktop/src/lib/keybinds/composer-focus-keys.test.ts
@@ -166,13 +166,14 @@ describe('composerFocusKeysAllowed', () => {
     expect(composerFocusKeysAllowed(keydown({ key: 'h', code: 'KeyH', target: document.body }), 'type')).toBe(true)
   })
 
-  it('refuses editables; refuses Enter on buttons but allows / and typing', () => {
+  it('refuses editables; leaves Enter and Space to buttons but allows / and typing', () => {
     const input = document.createElement('input')
     const button = document.createElement('button')
     document.body.append(input, button)
 
     expect(composerFocusKeysAllowed(keydown({ key: 'a', code: 'KeyA', target: input }), 'type')).toBe(false)
     expect(composerFocusKeysAllowed(keydown({ key: 'Enter', code: 'Enter', target: button }), 'enter')).toBe(false)
+    expect(composerFocusKeysAllowed(keydown({ key: ' ', code: 'Space', target: button }), 'type')).toBe(false)
     expect(composerFocusKeysAllowed(keydown({ key: '/', code: 'Slash', target: button }), '/')).toBe(true)
     expect(composerFocusKeysAllowed(keydown({ key: 'a', code: 'KeyA', target: button }), 'type')).toBe(true)
   })
diff --git a/apps/desktop/src/lib/keybinds/composer-focus-keys.ts b/apps/desktop/src/lib/keybinds/composer-focus-keys.ts
index 1c01f3ec4c..8c009991d8 100644
--- a/apps/desktop/src/lib/keybinds/composer-focus-keys.ts
+++ b/apps/desktop/src/lib/keybinds/composer-focus-keys.ts
@@ -184,5 +184,6 @@ export function composerFocusKeysAllowed(event: KeyboardEvent, combo: string): b
     return false
   }
 
-  return !(combo === 'enter' && isActivateOnEnterTarget(event.target))
+  // Space activates focused buttons too; it must not become a composer draft.
+  return !((combo === 'enter' || event.key === ' ') && isActivateOnEnterTarget(event.target))
 }
diff --git a/apps/desktop/src/store/composer-status-drawer.test.ts b/apps/desktop/src/store/composer-status-drawer.test.ts
new file mode 100644
index 0000000000..b7e0690673
--- /dev/null
+++ b/apps/desktop/src/store/composer-status-drawer.test.ts
@@ -0,0 +1,42 @@
+import { beforeEach, expect, it, vi } from 'vitest'
+
+beforeEach(() => {
+  window.localStorage.clear()
+  vi.resetModules()
+})
+
+it('restores the hidden choice without sharing it with another session or owner', async () => {
+  const store = await import('./composer-status-drawer')
+  const scope = { connectionId: 'local', profile: 'default', targetProfile: 'default', sessionId: 'chat-a' }
+  const key = store.statusDrawerKey(scope)
+  const otherSession = store.statusDrawerKey({ ...scope, sessionId: 'chat-b' })
+  const otherProfile = store.statusDrawerKey({ ...scope, profile: 'coder', targetProfile: 'coder' })
+  const otherConnection = store.statusDrawerKey({ ...scope, connectionId: 'remote' })
+
+  store.setStatusDrawerCollapsed(key, true)
+  store.setStatusDrawerCollapsed(otherSession, true)
+  store.setStatusDrawerCollapsed(otherSession, false)
+  vi.resetModules()
+
+  const restored = await import('./composer-status-drawer')
+  expect(restored.$collapsedStatusDrawers.get()).toEqual([key])
+  expect(restored.$collapsedStatusDrawers.get()).not.toContain(otherProfile)
+  expect(restored.$collapsedStatusDrawers.get()).not.toContain(otherConnection)
+})
+
+it('follows local profile renames and scopes profile removal to the exact owner', async () => {
+  const store = await import('./composer-status-drawer')
+  const scope = { connectionId: 'local', profile: 'coder', targetProfile: 'coder', sessionId: 'chat' }
+  const local = store.statusDrawerKey(scope)
+  const remote = store.statusDrawerKey({ ...scope, connectionId: 'remote' })
+  const renamed = store.statusDrawerKey({ ...scope, profile: 'work', targetProfile: 'work' })
+
+  store.setStatusDrawerCollapsed(local, true)
+  store.setStatusDrawerCollapsed(remote, true)
+  store.migrateStatusDrawersForProfile('coder', 'work')
+  expect(store.$collapsedStatusDrawers.get()).toEqual([renamed, remote])
+  store.dropStatusDrawersForProfile('work')
+  expect(store.$collapsedStatusDrawers.get()).toEqual([remote])
+  store.dropStatusDrawersForProfile('coder', { connectionId: 'remote', profile: 'coder', targetProfile: 'coder' })
+  expect(store.$collapsedStatusDrawers.get()).toEqual([])
+})
diff --git a/apps/desktop/src/store/composer-status-drawer.ts b/apps/desktop/src/store/composer-status-drawer.ts
new file mode 100644
index 0000000000..22c6035dd0
--- /dev/null
+++ b/apps/desktop/src/store/composer-status-drawer.ts
@@ -0,0 +1,100 @@
+import { Codecs, persistentAtom } from '@/lib/persisted'
+
+interface StatusDrawerScope {
+  connectionId: string | null
+  profile: string
+  targetProfile: string
+  sessionId: string
+}
+
+const MAX_SAVED_DRAWERS = 512
+
+function decodeScope(key: string): StatusDrawerScope | null {
+  try {
+    const value: unknown = JSON.parse(key)
+
+    if (
+      !Array.isArray(value) ||
+      value.length !== 4 ||
+      (value[0] !== null && typeof value[0] !== 'string') ||
+      !value.slice(1).every(part => typeof part === 'string' && part.length > 0)
+    ) {
+      return null
+    }
+
+    const [connectionId, profile, targetProfile, sessionId] = value as [string | null, string, string, string]
+
+    return { connectionId, profile, targetProfile, sessionId }
+  } catch {
+    return null
+  }
+}
+
+export function statusDrawerKey(scope: StatusDrawerScope): string {
+  return JSON.stringify([scope.connectionId, scope.profile, scope.targetProfile, scope.sessionId])
+}
+
+/** Only hidden drawers need an entry; new conversations keep the existing open default. */
+export const $collapsedStatusDrawers = persistentAtom<string[]>(
+  'hermes.desktop.collapsedStatusDrawers.v1',
+  [],
+  Codecs.json(value =>
+    Array.isArray(value)
+      ? [...new Set(value.filter((key): key is string => typeof key === 'string' && decodeScope(key) !== null))].slice(
+          -MAX_SAVED_DRAWERS
+        )
+      : []
+  )
+)
+
+export function setStatusDrawerCollapsed(key: string, collapsed: boolean): void {
+  const next = $collapsedStatusDrawers.get().filter(saved => saved !== key)
+
+  if (collapsed) {
+    next.push(key)
+  }
+
+  $collapsedStatusDrawers.set(next.slice(-MAX_SAVED_DRAWERS))
+}
+
+export function migrateStatusDrawersForProfile(from: string, to: string): void {
+  const renamed = $collapsedStatusDrawers.get().map(key => {
+    const scope = decodeScope(key)
+
+    if (!scope || (scope.connectionId !== null && scope.connectionId !== 'local')) {
+      return key
+    }
+
+    return statusDrawerKey({
+      ...scope,
+      profile: scope.profile === from ? to : scope.profile,
+      targetProfile: scope.targetProfile === from ? to : scope.targetProfile
+    })
+  })
+
+  $collapsedStatusDrawers.set([...new Set(renamed)])
+}
+
+export function dropStatusDrawersForProfile(
+  profile: string,
+  route?: { connectionId?: string; profile?: string; targetProfile?: string }
+): void {
+  $collapsedStatusDrawers.set(
+    $collapsedStatusDrawers.get().filter(key => {
+      const scope = decodeScope(key)
+
+      if (!scope) {
+        return false
+      }
+
+      const matches = route
+        ? scope.profile === route.profile?.trim() &&
+          (!route.targetProfile || scope.targetProfile === route.targetProfile.trim()) &&
+          (!route.connectionId || scope.connectionId === route.connectionId.trim())
+        : (scope.connectionId === null || scope.connectionId === 'local') &&
+          (scope.profile === profile || scope.targetProfile === profile)
+
+      return !matches
+    })
+  )
+}
diff --git a/apps/desktop/src/store/session-states.ts b/apps/desktop/src/store/session-states.ts
index 95e56af19d..ad815c6b80 100644
--- a/apps/desktop/src/store/session-states.ts
+++ b/apps/desktop/src/store/session-states.ts
@@ -36,6 +36,7 @@ import { stableArray } from '@/lib/stable-array'
 import { readJson, writeJson } from '@/lib/storage'
 import type { SessionInfo } from '@/types/hermes'
 
+import { dropStatusDrawersForProfile, migrateStatusDrawersForProfile } from './composer-status-drawer'
 import { dropPreviewTabsForProfile, migratePreviewTabsForProfile, setPreviewScope } from './preview'
 import { dropPreviewArtifactsForProfile, migratePreviewArtifactsForProfile } from './preview-status'
 import { $activeGatewayProfile, normalizeProfileKey } from './profile'
@@ -2080,6 +2081,7 @@ export function dropTilesForProfile(
 
   const name = normalizeProfileKey(profile)
   dropPreviewArtifactsForProfile(name, route)
+  dropStatusDrawersForProfile(name, route)
   // Route fields go through the SAME canonicalization as `name` below — a
   // source-scoped delete must not be defeated by stray whitespace around a
   // profile name that a non-route delete trims away.
@@ -2222,6 +2224,7 @@ export function migrateTilesForProfile(oldProfile: string, newProfile: string):
   migrateRememberedNavigationForProfile(from, to)
   migrateSessionOwnerHintsForProfile(from, to)
   migratePreviewArtifactsForProfile(from, to)
+  migrateStatusDrawersForProfile(from, to)
   // Sibling family: the rail's profile-keyed buckets move with the rename, or
   // the renamed profile opens with an empty rail and the old name keeps them.
   migratePreviewTabsForProfile(from, to)
diff --git a/gateway/run_notifications.py b/gateway/run_notifications.py
index 56999f56e8..a9aa431b64 100644
--- a/gateway/run_notifications.py
+++ b/gateway/run_notifications.py
@@ -14,6 +14,7 @@ import json
 import logging
 import time
 from contextlib import suppress
+from datetime import datetime
 from pathlib import Path
 from typing import Any, Dict, Optional, cast
 
@@ -32,6 +33,12 @@ _UPDATE_FAILED_NOTICE = (
     "❌ Hermes update failed; the previous version is still running. Run `hermes update` on the "
     "host to see the full error, or try /update again later.")
 
+# An update's completion notice waits for its target platform adapter to (re)connect before it
+# can be delivered. Nothing bounds that wait, so a marker naming a platform that is not
+# configured at all — no adapter will ever appear — would keep itself on disk and re-log a
+# deferred line on every poll, in every process, forever. Stop waiting past this age.
+_UPDATE_NOTIFY_MAX_ADAPTER_WAIT_SECONDS = 3600.0
+
 
 def _served_notice_target_key(profile: Optional[str], platform_value: str, chat_id, thread_id) -> tuple:
     """Notice-dedupe key for one SERVED profile's home channel.
@@ -522,6 +529,24 @@ class GatewayNotificationsMixin:
             return profile_from_session_key_namespace(parts[1])
         return None
 
+    @staticmethod
+    def _marker_age_seconds(data: dict) -> Optional[float]:
+        """Age of a persisted update marker, from the ``timestamp`` stamped by its writer.
+
+        ``None`` when the marker carries no parseable stamp — the field is absent on markers
+        written before it existed, and callers keep the old retry behavior rather than guess.
+        """
+        raw = str(data.get("timestamp") or "").strip()
+        if not raw:
+            return None
+        try:
+            stamped = datetime.fromisoformat(raw)
+        except ValueError:
+            return None
+        # The writer stamps a naive local ``datetime.now()``; tolerate a tz-aware one too.
+        now = datetime.now(stamped.tzinfo) if stamped.tzinfo else datetime.now()
+        return (now - stamped).total_seconds()
+
     def _resolve_update_target(self, paths: "_UpdatePaths") -> Optional["_UpdateTarget"]:
         """Resolve adapter/chat/session for update watcher messages from the pending marker."""
         for path in (paths.claimed, paths.pending):
@@ -731,6 +756,18 @@ class GatewayNotificationsMixin:
             platform = Platform(platform_str)
             adapter = self._authorization_adapter(platform, self._marker_profile(pending))
             if chat_id and not adapter:
+                age = self._marker_age_seconds(pending)
+                if age is not None and age > _UPDATE_NOTIFY_MAX_ADAPTER_WAIT_SECONDS:
+                    # The platform never came back. Deferring forever leaks the markers and re-logs
+                    # on every poll for the life of the install: the startup path reschedules this
+                    # watcher whenever the markers are still on disk, so an undeliverable marker
+                    # outlives every restart. Give up loudly, clear the markers, and report a
+                    # definitive decision (True) so the caller stops rescheduling.
+                    logger.warning(
+                        "Post-update notification for %s:%s dropped after %.1fh: %s adapter never "
+                        "connected", platform_str, chat_id, age / 3600.0, platform_str)
+                    self._clear_update_markers(paths, pending.get("session_key"))
+                    return True
                 # Target platform not reconnected yet (common right after the update's restart): keep the
                 # markers for a later retry instead of silently losing the notification.
                 return _defer("Update notification deferred: %s adapter not connected yet", platform_str)
diff --git a/hermes_cli/models_catalog_static.py b/hermes_cli/models_catalog_static.py
index fe79673652..d468b9c760 100644
--- a/hermes_cli/models_catalog_static.py
+++ b/hermes_cli/models_catalog_static.py
@@ -159,7 +159,10 @@ _XAI_MODELS = _xai_curated_models()
 # Curated per-provider lists. ``-cn`` twins share the international catalog on a domestic endpoint.
 _PROVIDER_MODELS: dict[str, list[str]] = {
     "moa": ["default"],
-    "nous": [mid for mid, _ in OPENROUTER_MODELS if mid not in _OPENROUTER_ONLY and not mid.endswith(":free")],
+    # grok-4.7 is on the Nous Portal ahead of the OpenRouter listing; pinned first so it heads the picker.
+    "nous": ["x-ai/grok-4.7"] + [
+        mid for mid, _ in OPENROUTER_MODELS if mid not in _OPENROUTER_ONLY and not mid.endswith(":free")
+    ],
     # Used by /model counts and provider_model_ids fallback when /v1/models is unavailable.
     "openai": list(_OPENAI_CHAT_MODELS),
     "openai-api": [
diff --git a/plugins/video_gen/fal/__init__.py b/plugins/video_gen/fal/__init__.py
index f79df81b04..11c8d06420 100644
--- a/plugins/video_gen/fal/__init__.py
+++ b/plugins/video_gen/fal/__init__.py
@@ -274,10 +274,15 @@ def _submit_fal_video_request(endpoint: str, arguments: Dict[str, Any]):
     managed_gateway = _resolve_managed_fal_video_gateway()
     if managed_gateway is None:
         return client.submit(endpoint, arguments=arguments, headers=headers)
+    from tools.fal_common import (
+        _extract_http_status, _managed_fal_billing_error, submit_managed_fal_with_rate_limit_retry,
+    )
     try:
-        return _get_managed_fal_video_client(managed_gateway).submit(endpoint, arguments=arguments, headers=headers)
+        return submit_managed_fal_with_rate_limit_retry(
+            lambda request_headers: _get_managed_fal_video_client(managed_gateway).submit(
+                endpoint, arguments=arguments, headers=request_headers),
+            what="video endpoint", name=endpoint)
     except Exception as exc:
-        from tools.fal_common import _extract_http_status, _managed_fal_billing_error
         status = _extract_http_status(exc)
         if status is not None and 400 <= status < 500:
             billing = _managed_fal_billing_error(exc, "endpoint")
diff --git a/tests/gateway/test_update_command.py b/tests/gateway/test_update_command.py
index bc858f9a96..cb9cdab382 100644
--- a/tests/gateway/test_update_command.py
+++ b/tests/gateway/test_update_command.py
@@ -5,6 +5,7 @@ the _send_update_notification startup hook (sends results after restart).
 """
 
 import json
+from datetime import datetime, timedelta
 from pathlib import Path
 from unittest.mock import patch, MagicMock, AsyncMock
 
@@ -366,6 +367,68 @@ class TestSendUpdateNotification:
         assert "Update complete" in call_args[0][1] or "update finished" in call_args[0][1].lower()
 
 
+    @pytest.mark.asyncio
+    async def test_drops_stale_marker_when_the_platform_never_connects(self, tmp_path, caplog):
+        """A marker past the wait cap is abandoned instead of deferred forever.
+
+        Regression: an update notice addressed to a platform that has no adapter —
+        and never will, because the platform is not configured at all — kept its
+        markers on disk and re-logged a deferred line on every poll. The startup
+        path reschedules the watcher for as long as the markers exist, so the
+        notice outlived every restart, in every process.
+        """
+        runner = _make_runner()
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+
+        pending_path = hermes_home / ".update_pending.json"
+        pending_path.write_text(json.dumps({
+            "platform": "telegram",
+            "chat_id": "67890",
+            "user_id": "12345",
+            "timestamp": (datetime.now() - timedelta(hours=2)).isoformat(),
+        }))
+        (hermes_home / ".update_exit_code").write_text("0")
+        # runner.adapters stays empty: no adapter for the target platform, ever.
+
+        with patch("gateway.run._hermes_home", hermes_home):
+            result = await runner._send_update_notification()
+
+        # True is the definitive answer the startup caller keys off to stop rescheduling.
+        assert result is True
+        assert not pending_path.exists()
+        assert not (hermes_home / ".update_pending.claimed.json").exists()
+        assert not (hermes_home / ".update_output.txt").exists()
+        assert not (hermes_home / ".update_exit_code").exists()
+        assert any("adapter never connected" in r.getMessage() for r in caplog.records)
+
+    @pytest.mark.asyncio
+    async def test_keeps_waiting_for_a_recent_marker(self, tmp_path):
+        """A recent marker is still held: the cap must not swallow its own notice.
+
+        Right after the update's restart the adapter is legitimately absent for a
+        while, which is the case the defer path exists to cover.
+        """
+        runner = _make_runner()
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+
+        pending_path = hermes_home / ".update_pending.json"
+        pending_path.write_text(json.dumps({
+            "platform": "telegram",
+            "chat_id": "67890",
+            "user_id": "12345",
+            "timestamp": (datetime.now() - timedelta(minutes=5)).isoformat(),
+        }))
+        (hermes_home / ".update_exit_code").write_text("0")
+
+        with patch("gateway.run._hermes_home", hermes_home):
+            result = await runner._send_update_notification()
+
+        assert result is False
+        assert pending_path.exists(), "marker kept so a later poll can still deliver it"
+
+
     @pytest.mark.asyncio
     async def test_cleans_up_on_error(self, tmp_path):
         """Files are cleaned up even if notification fails."""
diff --git a/tests/tools/test_image_generation.py b/tests/tools/test_image_generation.py
index ddc6bfd556..4971b92183 100644
--- a/tests/tools/test_image_generation.py
+++ b/tests/tools/test_image_generation.py
@@ -563,6 +563,42 @@ class TestManagedGatewayErrorTranslation:
         with pytest.raises(ConnectionError):
             image_tool._submit_fal_request("fal-ai/flux-2-pro", {"prompt": "x"})
 
+    @staticmethod
+    def _rate_limited(retry_after):
+        return _MockHttpxError(429, "Too Many Requests", payload={"error": {
+            "code": "RATE_LIMIT_EXCEEDED", "message": "Rate limit exceeded.", "retryAfter": retry_after}})
+
+    def test_short_429_is_retried_once_under_a_fresh_idempotency_key(self, image_tool, monkeypatch):
+        """A gateway 429 with a short retryAfter is waited out and resubmitted once (new key)."""
+        from unittest.mock import MagicMock
+
+        monkeypatch.setattr(image_tool, "_resolve_managed_fal_gateway", lambda: MagicMock())
+        mock_managed_client = MagicMock()
+        mock_managed_client.submit.side_effect = [self._rate_limited(1), "handle"]
+        monkeypatch.setattr(image_tool, "_get_managed_fal_client", lambda gw: mock_managed_client)
+
+        assert image_tool._submit_fal_request("fal-ai/gpt-image-2", {"prompt": "x"}) == "handle"
+
+        keys = [call.kwargs["headers"]["x-idempotency-key"] for call in mock_managed_client.submit.call_args_list]
+        assert len(keys) == 2 and keys[0] != keys[1]
+
+    def test_long_429_is_reported_as_a_rate_limit_not_a_missing_model(self, image_tool, monkeypatch):
+        """A 429 beyond the retry cap names the rate limit; agents must not be told to switch models."""
+        from unittest.mock import MagicMock
+
+        monkeypatch.setattr(image_tool, "_resolve_managed_fal_gateway", lambda: MagicMock())
+        mock_managed_client = MagicMock()
+        mock_managed_client.submit.side_effect = self._rate_limited(120)
+        monkeypatch.setattr(image_tool, "_get_managed_fal_client", lambda gw: mock_managed_client)
+
+        with pytest.raises(ValueError) as exc_info:
+            image_tool._submit_fal_request("fal-ai/gpt-image-2", {"prompt": "x"})
+
+        msg = str(exc_info.value)
+        assert "rate-limited" in msg and "retry after 120s" in msg
+        assert "may not yet be enabled" not in msg
+        assert mock_managed_client.submit.call_count == 1
+
 
 class TestKreaModelNormalization:
     """Native ``krea-2-*`` detection for managed Krea routing."""
diff --git a/tools/fal_common.py b/tools/fal_common.py
index c7803e4a5b..41601ce511 100644
--- a/tools/fal_common.py
+++ b/tools/fal_common.py
@@ -26,9 +26,14 @@ issue #26241 for details.
 
 from __future__ import annotations
 
-from typing import Any, Dict, Optional, Union
+import time
+import uuid
+from typing import Any, Callable, Dict, Optional, Union
 from urllib.parse import urlencode
 
+# A 429 whose Retry-After is longer than this is reported, not waited out inside a tool call.
+MANAGED_FAL_RATE_LIMIT_RETRY_CAP_SECONDS = 30.0
+
 
 def import_fal_client() -> Any:
     """Import ``fal_client`` (via ``pm`` when available) and return
@@ -99,6 +104,61 @@ def _managed_fal_billing_error(exc: BaseException, what: str) -> Optional[str]:
     )
 
 
+def _managed_fal_retry_after_seconds(exc: BaseException) -> Optional[float]:
+    """Seconds the managed gateway asked us to wait after a 429: the ``Retry-After`` header,
+    else the body's ``error.retryAfter``; None when the status is not 429 or neither is present."""
+    response = getattr(exc, "response", None)
+    if response is None or _extract_http_status(exc) != 429:
+        return None
+    headers = getattr(response, "headers", None)
+    raw = headers.get("Retry-After") if headers is not None and hasattr(headers, "get") else None
+    if raw is None:
+        try:
+            error = response.json().get("error")
+        except Exception:  # noqa: BLE001 — a non-JSON 429 body simply has no hint
+            return None
+        raw = error.get("retryAfter") if isinstance(error, dict) else None
+    try:
+        return float(raw) if raw is not None else None
+    except (TypeError, ValueError):
+        return None
+
+
+def _managed_fal_rate_limit_message(what: str, name: str, retry_after: Optional[float]) -> str:
+    hint = f"retry after {retry_after:g}s" if retry_after is not None else "no Retry-After given"
+    return (
+        f"Nous Subscription gateway rate-limited {what} '{name}' (HTTP 429; {hint}). "
+        "The model is enabled — retry later instead of switching models or setting FAL_KEY."
+    )
+
+
+def submit_managed_fal_with_rate_limit_retry(
+    submit: Callable[[Dict[str, str]], Any], *, what: str, name: str,
+):
+    """Call ``submit(headers)`` with a fresh ``x-idempotency-key``; on a 429 whose Retry-After
+    fits the cap, wait it out (interrupt-aware) and resubmit ONCE under a new key.
+
+    A second 429, or one with an unknown/too-long Retry-After, raises ValueError naming the
+    rate limit — it must never fall through to the callers' "model may not be enabled" 4xx text,
+    which sends agents off to switch models. Every other exception propagates untouched.
+    """
+    from tools.interrupt import is_interrupted
+    for attempt in (1, 2):
+        try:
+            return submit({"x-idempotency-key": str(uuid.uuid4())})
+        except Exception as exc:
+            if _extract_http_status(exc) != 429:
+                raise
+            retry_after = _managed_fal_retry_after_seconds(exc)
+            if attempt == 2 or retry_after is None or retry_after > MANAGED_FAL_RATE_LIMIT_RETRY_CAP_SECONDS:
+                raise ValueError(_managed_fal_rate_limit_message(what, name, retry_after)) from exc
+            deadline = time.monotonic() + retry_after
+            while time.monotonic() < deadline:
+                if is_interrupted():
+                    raise ValueError(_managed_fal_rate_limit_message(what, name, retry_after)) from exc
+                time.sleep(min(0.5, max(0.0, deadline - time.monotonic())))
+
+
 def _require(value: Any, what: str) -> Any:
     if value is None:
         raise RuntimeError(f"{what} is required for managed FAL gateway mode")
diff --git a/tools/image_generation_tool.py b/tools/image_generation_tool.py
index 6d00b89fcd..6c7d684118 100644
--- a/tools/image_generation_tool.py
+++ b/tools/image_generation_tool.py
@@ -31,7 +31,7 @@ def _load_fal_client() -> Any:
 from tools.debug_helpers import DebugSession
 from tools.fal_common import (
     _ManagedFalSyncClient, _extract_http_status, _managed_fal_billing_error,
-    _normalize_fal_queue_url_format,
+    _normalize_fal_queue_url_format, submit_managed_fal_with_rate_limit_retry,
 )
 from tools.image_generation_catalog import (
     DEFAULT_ASPECT_RATIO, DEFAULT_MODEL, FAL_MODELS, UPSCALER_CREATIVITY, UPSCALER_DEFAULT_PROMPT,
@@ -127,8 +127,10 @@ def _submit_fal_request(model: str, arguments: Dict[str, Any]):
     if managed_gateway is None:
         return fal_client.submit(model, arguments=arguments, headers=request_headers)
     try:
-        return _get_managed_fal_client(managed_gateway).submit(
-            model, arguments=arguments, headers=request_headers)
+        return submit_managed_fal_with_rate_limit_retry(
+            lambda headers: _get_managed_fal_client(managed_gateway).submit(
+                model, arguments=arguments, headers=headers),
+            what="image model", name=model)
     except Exception as exc:
         # A managed-gateway 4xx usually means the portal doesn't proxy this model
         # (allowlist miss, billing gate): give remediation instead of a raw httpx error.
diff --git a/website/static/api/model-catalog.json b/website/static/api/model-catalog.json
index d35465ad42..fc1ad48036 100644
--- a/website/static/api/model-catalog.json
+++ b/website/static/api/model-catalog.json
@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "updated_at": "2026-09-20T18:57:41Z",
+  "updated_at": "2026-09-21T17:31:22Z",
   "metadata": {
     "source": "hermes-agent repo",
     "docs": "https://hermes-agent.nousresearch.com/docs/reference/model-catalog"
@@ -397,4 +397,4 @@
       ]
     }
   }
-}
\ No newline at end of file
+}