feat(desktop): add desktop.reset_model_on_new_session config flag

apps/desktop/src/app/desktop-controller.tsx

@@ -1017,8 +1017,11 @@ export function DesktopController() {
 
   useEffect(() => {
     if (gatewayState === 'open' && !activeSessionId && freshDraftReady) {
-      void refreshCurrentModel()
-      void refreshHermesConfig()
+      // Load config first so the reset-on-new-session flag is in the store
+      // before refreshCurrentModel decides whether to force-reseed the model.
+      void refreshHermesConfig().finally(() => {
+        void refreshCurrentModel()
+      })
     }
   }, [activeSessionId, freshDraftReady, gatewayState, refreshCurrentModel, refreshHermesConfig])
 

apps/desktop/src/app/session/hooks/use-hermes-config.test.ts

@@ -4,7 +4,7 @@ import { beforeEach, describe, expect, it, vi } from 'vitest'
 
 import { getHermesConfig } from '@/hermes'
 import { persistString } from '@/lib/storage'
-import { $currentCwd, setCurrentCwd } from '@/store/session'
+import { $currentCwd, $resetModelOnNewSession, setCurrentCwd, setResetModelOnNewSession } from '@/store/session'
 
 import { useHermesConfig } from './use-hermes-config'
 
@@ -22,6 +22,7 @@ describe('useHermesConfig refreshHermesConfig', () => {
   beforeEach(() => {
     // Reset atoms and localStorage between tests
     setCurrentCwd('')
+    setResetModelOnNewSession(false)
     persistString(WORKSPACE_CWD_KEY, null)
   })
 
@@ -141,4 +142,39 @@ describe('useHermesConfig refreshHermesConfig', () => {
 
     expect(refreshProjectBranch).toHaveBeenCalledWith('/workspace/attached-project')
   })
+
+  it('mirrors desktop.reset_model_on_new_session=true into the store', async () => {
+    mockConfig({ desktop: { reset_model_on_new_session: true } })
+
+    const { result } = renderHook(() =>
+      useHermesConfig({
+        activeSessionIdRef: { current: null },
+        refreshProjectBranch: vi.fn().mockResolvedValue(undefined)
+      })
+    )
+
+    await act(async () => {
+      await result.current.refreshHermesConfig()
+    })
+
+    expect($resetModelOnNewSession.get()).toBe(true)
+  })
+
+  it('defaults the reset-model flag to false when the config key is absent', async () => {
+    setResetModelOnNewSession(true) // prove the refresh clears a stale true
+    mockConfig({})
+
+    const { result } = renderHook(() =>
+      useHermesConfig({
+        activeSessionIdRef: { current: null },
+        refreshProjectBranch: vi.fn().mockResolvedValue(undefined)
+      })
+    )
+
+    await act(async () => {
+      await result.current.refreshHermesConfig()
+    })
+
+    expect($resetModelOnNewSession.get()).toBe(false)
+  })
 })

apps/desktop/src/app/session/hooks/use-hermes-config.ts

@@ -11,7 +11,8 @@ import {
   setCurrentPersonality,
   setCurrentReasoningEffort,
   setCurrentServiceTier,
-  setIntroPersonality
+  setIntroPersonality,
+  setResetModelOnNewSession
 } from '@/store/session'
 import { applyAutoSpeakFromConfig } from '@/store/voice-prefs'
 
@@ -87,6 +88,7 @@ export function useHermesConfig({ activeSessionIdRef, refreshProjectBranch }: He
 
       setVoiceMaxRecordingSeconds(recordingLimit(config.voice?.max_recording_seconds))
       setSttEnabled(config.stt?.enabled !== false)
+      setResetModelOnNewSession(config.desktop?.reset_model_on_new_session === true)
       applyAutoSpeakFromConfig(config)
     } catch {
       // Config is nice-to-have; chat still works without it.

apps/desktop/src/app/session/hooks/use-model-controls.test.tsx

@@ -3,7 +3,14 @@ import { cleanup, render, renderHook } from '@testing-library/react'
 import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 
 import { getGlobalModelInfo } from '@/hermes'
-import { $activeSessionId, $currentModel, $currentProvider, setCurrentModel, setCurrentProvider } from '@/store/session'
+import {
+  $activeSessionId,
+  $currentModel,
+  $currentProvider,
+  setCurrentModel,
+  setCurrentProvider,
+  setResetModelOnNewSession
+} from '@/store/session'
 
 import { useModelControls } from './use-model-controls'
 
@@ -56,6 +63,7 @@ describe('useModelControls', () => {
     $activeSessionId.set(null)
     setCurrentModel('')
     setCurrentProvider('')
+    setResetModelOnNewSession(false)
   })
 
   afterEach(() => {
@@ -64,6 +72,7 @@ describe('useModelControls', () => {
     $activeSessionId.set(null)
     setCurrentModel('')
     setCurrentProvider('')
+    setResetModelOnNewSession(false)
   })
 
   it('applies the global model when there is no active runtime session', async () => {
@@ -179,4 +188,46 @@ describe('useModelControls', () => {
     await result.current.refreshCurrentModel(true)
     expect($currentModel.get()).toBe('openai/gpt-5.5')
   })
+
+  it('reseeds a sticky last-picked model to the default when reset_model_on_new_session is ON', async () => {
+    vi.mocked(getGlobalModelInfo).mockResolvedValue({ model: 'claude-opus-4-8', provider: 'claude-apr' })
+    setCurrentModel('claude-fable-5')
+    setCurrentProvider('claude-apr')
+    setResetModelOnNewSession(true)
+
+    const { result } = renderHook(() =>
+      useModelControls({
+        activeSessionId: null,
+        queryClient: new QueryClient(),
+        requestGateway: vi.fn()
+      })
+    )
+
+    // Flag ON: a fresh draft snaps the sticky pick back to the profile default.
+    await result.current.refreshCurrentModel()
+    expect($currentModel.get()).toBe('claude-opus-4-8')
+    expect($currentProvider.get()).toBe('claude-apr')
+  })
+
+  it('never reseeds the composer for an ACTIVE session even when the flag is ON', async () => {
+    vi.mocked(getGlobalModelInfo).mockResolvedValue({ model: 'claude-opus-4-8', provider: 'claude-apr' })
+    setCurrentModel('claude-fable-5')
+    setCurrentProvider('claude-apr')
+    setResetModelOnNewSession(true)
+    $activeSessionId.set('runtime-1')
+
+    const { result } = renderHook(() =>
+      useModelControls({
+        activeSessionId: 'runtime-1',
+        queryClient: new QueryClient(),
+        requestGateway: vi.fn()
+      })
+    )
+
+    await result.current.refreshCurrentModel()
+
+    // A live session owns its model; the flag must not disturb it mid-conversation.
+    expect($currentModel.get()).toBe('claude-fable-5')
+    expect($currentProvider.get()).toBe('claude-apr')
+  })
 })

apps/desktop/src/app/session/hooks/use-model-controls.ts

@@ -4,7 +4,7 @@ import { useCallback } from 'react'
 import { getGlobalModelInfo } from '@/hermes'
 import { useI18n } from '@/i18n'
 import { notifyError } from '@/store/notifications'
-import { $activeSessionId, $currentModel, $currentProvider, setCurrentModel, setCurrentProvider } from '@/store/session'
+import { $activeSessionId, $currentModel, $currentProvider, $resetModelOnNewSession, setCurrentModel, setCurrentProvider } from '@/store/session'
 import type { ModelOptionsResponse } from '@/types/hermes'
 
 interface ModelSelection {
@@ -40,19 +40,26 @@ export function useModelControls({ activeSessionId, queryClient, requestGateway
   // only fills an EMPTY selection so a user's pick (plain UI state in
   // $currentModel) survives the lifecycle refreshes that fire on boot / fresh
   // draft / session events. A live session owns the footer, so skip entirely.
+  //
+  // When config.yaml `desktop.reset_model_on_new_session` is on, a fresh draft
+  // is treated as a forced reseed too: the sticky last-picked model is
+  // overwritten with the profile default so a new chat always starts on the
+  // default (opt-in; default off preserves the sticky behavior).
   const refreshCurrentModel = useCallback(async (force = false) => {
     try {
       if ($activeSessionId.get()) {
         return
       }
 
-      if (!force && $currentModel.get()) {
+      const effectiveForce = force || $resetModelOnNewSession.get()
+
+      if (!effectiveForce && $currentModel.get()) {
         return
       }
 
       const result = await getGlobalModelInfo()
 
-      if ($activeSessionId.get() || (!force && $currentModel.get())) {
+      if ($activeSessionId.get() || (!effectiveForce && $currentModel.get())) {
         return
       }
 

apps/desktop/src/store/session.ts

@@ -265,6 +265,11 @@ export const $currentProvider = atom(storedString(COMPOSER_PROVIDER_KEY) ?? '')
 export const $currentReasoningEffort = atom(storedString(COMPOSER_EFFORT_KEY) ?? '')
 export const $currentServiceTier = atom('')
 export const $currentFastMode = atom(storedBoolean(COMPOSER_FAST_KEY, false))
+// Mirror of config.yaml `desktop.reset_model_on_new_session` (default false).
+// When true, a fresh chat reseeds the composer model to the profile default
+// instead of carrying the last-picked model forward. Refreshed from backend
+// config by useHermesConfig; a plain reflection, not its own persisted state.
+export const $resetModelOnNewSession = atom(false)
 // Effective approval-bypass state mirrored from the gateway (session.info).
 // Persistence lives in the backend config (approvals.mode), so this is a plain
 // reflection of the truth the gateway reports rather than its own store.
@@ -331,6 +336,9 @@ export const setCurrentFastMode = (next: Updater<boolean>) => {
   persistBoolean(COMPOSER_FAST_KEY, $currentFastMode.get())
 }
 
+export const setResetModelOnNewSession = (next: Updater<boolean>) =>
+  updateAtom($resetModelOnNewSession, next)
+
 export const setYoloActive = (next: Updater<boolean>) => updateAtom($yoloActive, next)
 
 export const setCurrentCwd = (next: Updater<string>) => {

apps/desktop/src/types/hermes.ts

@@ -220,6 +220,12 @@ export interface HermesConfig {
     max_recording_seconds?: number
     auto_tts?: boolean
   }
+  desktop?: {
+    // When true, a fresh chat (Cmd+N / app relaunch) reseeds the composer model
+    // to the profile default instead of carrying the last-picked model forward.
+    // Default false preserves the sticky-last-pick behavior.
+    reset_model_on_new_session?: boolean
+  }
 }
 
 export type HermesConfigRecord = Record<string, unknown>