fix(desktop): scope model options by profile (#62795)

Co-authored-by: embwl0x <embwl0x@users.noreply.github.com>

apps/desktop/src/app/chat/index.tsx

@@ -14,18 +14,19 @@ import { PromptOverlays } from '@/components/prompt-overlays'
 import { Button } from '@/components/ui/button'
 import { ErrorState } from '@/components/ui/error-state'
 import { TitleMenuTrigger } from '@/components/ui/title-menu-trigger'
-import { getGlobalModelOptions, type HermesGateway } from '@/hermes'
+import { type HermesGateway } from '@/hermes'
 import { useI18n } from '@/i18n'
 import type { ChatMessage } from '@/lib/chat-messages'
 import { quickModelOptions, sessionTitle } from '@/lib/chat-runtime'
 import { useIncrementalExternalStoreRuntime } from '@/lib/incremental-external-store-runtime'
+import { modelOptionsQueryKey, requestModelOptions } from '@/lib/model-options'
 import { cn } from '@/lib/utils'
 import { migrateSessionDraft } from '@/store/composer'
 import { migrateQueuedPrompts, parkQueuedPrompts } from '@/store/composer-queue'
 import { $pinnedSessionIds } from '@/store/layout'
 import { $petActive } from '@/store/pet'
 import { $petOverlayActive } from '@/store/pet-overlay'
-import { $gatewaySwapTarget, $profiles } from '@/store/profile'
+import { $activeGatewayProfile, $gatewaySwapTarget, $profiles } from '@/store/profile'
 import {
   $contextSuggestions,
   $freshDraftReady,
@@ -254,6 +255,7 @@ export function ChatView({
   const sessionAnchor = isPrimary ? 'workspace' : `session-tile:${storedId ?? ''}`
   const awaitingResponse = useStore(view.$awaitingResponse)
   const busy = useStore(view.$busy)
+  const activeGatewayProfile = useStore($activeGatewayProfile)
   const contextSuggestions = useStore($contextSuggestions)
   // Per-session (SessionView) reads — a tile IS its session, so these come
   // from the view slice, not the global atoms (which track the primary only).
@@ -356,21 +358,8 @@ export function ChatView({
   const threadKey = selectedSessionId || activeSessionId || (isRoutedSessionView ? location.pathname : 'new')
 
   const modelOptionsQuery = useQuery<ModelOptionsResponse>({
-    queryKey: ['model-options', activeSessionId || 'global'],
-    queryFn: () => {
-      if (!activeSessionId) {
-        return getGlobalModelOptions()
-      }
-
-      if (!gateway) {
-        throw new Error('Hermes gateway unavailable')
-      }
-
-      return gateway.request<ModelOptionsResponse>('model.options', {
-        session_id: activeSessionId,
-        explicit_only: true
-      })
-    },
+    queryKey: modelOptionsQueryKey(activeGatewayProfile, activeSessionId),
+    queryFn: () => requestModelOptions({ gateway: gateway || undefined, sessionId: activeSessionId }),
     enabled: gatewayOpen
   })
 

apps/desktop/src/app/chat/session-tile.tsx

@@ -35,6 +35,7 @@ import type { ChatMessage } from '@/lib/chat-messages'
 import { sessionTitle } from '@/lib/chat-runtime'
 import { createComposerAttachmentScope } from '@/store/composer'
 import { $pinnedSessionIds, pinSession, unpinSession } from '@/store/layout'
+import { $activeGatewayProfile } from '@/store/profile'
 import { sessionAwaitingInput } from '@/store/prompts'
 import {
   $gatewayState,
@@ -112,6 +113,7 @@ function TileChat({
   const { gatewayRef, requestGateway } = useGatewayRequest()
   const queryClient = useQueryClient()
   const { selectModel } = useModelControls({ queryClient, requestGateway })
+  const activeGatewayProfile = useStore($activeGatewayProfile)
   const cwd = useStore(view.$cwd)
   const gatewayOpen = useStore($gatewayState) === 'open'
 
@@ -148,10 +150,11 @@ function TileChat({
         <ModelMenuPanel
           gateway={gatewayRef.current || undefined}
           onSelectModel={selectModel}
+          profile={activeGatewayProfile}
           requestGateway={requestGateway}
         />
       ) : null,
-    [gatewayOpen, gatewayRef, requestGateway, selectModel]
+    [activeGatewayProfile, gatewayOpen, gatewayRef, requestGateway, selectModel]
   )
 
   return (

apps/desktop/src/app/contrib/surfaces.tsx

@@ -13,6 +13,7 @@ import { Navigate, Route, Routes, useParams } from 'react-router-dom'
 
 import { ContribBoundary } from '@/contrib/react/boundary'
 import { useContributions } from '@/contrib/react/use-contributions'
+import { $activeGatewayProfile } from '@/store/profile'
 import { $freshDraftReady, $gatewayState } from '@/store/session'
 
 import { ChatView } from '../chat'
@@ -109,6 +110,7 @@ export const ChatRoutesSurface = memo(function ChatRoutesSurface({
   actions: WiringActions
   maxVoiceRecordingSeconds?: number
 }) {
+  const activeGatewayProfile = useStore($activeGatewayProfile)
   const gatewayState = useStore($gatewayState)
   useContributions(ROUTES_AREA)
   const routeContributions = contributedRoutes()
@@ -128,10 +130,11 @@ export const ChatRoutesSurface = memo(function ChatRoutesSurface({
         <ModelMenuPanel
           gateway={gateway || undefined}
           onSelectModel={actions.selectModel}
+          profile={activeGatewayProfile}
           requestGateway={actions.requestGateway}
         />
       ) : null,
-    [actions, gateway, gatewayState]
+    [actions, activeGatewayProfile, gateway, gatewayState]
   )
 
   const chatView = (

apps/desktop/src/app/contrib/wiring.tsx

@@ -139,6 +139,7 @@ export function ContribWiring({ children }: { children: ReactNode }) {
   const resumeExhaustedSessionId = useStore($resumeExhaustedSessionId)
   const selectedStoredSessionId = useStore($selectedStoredSessionId)
   const messagingSessions = useStore($messagingSessions)
+  const activeGatewayProfile = useStore($activeGatewayProfile)
   const profileScope = useStore($profileScope)
 
   const routedSessionId = routeSessionId(location.pathname)
@@ -341,6 +342,7 @@ export function ContribWiring({ children }: { children: ReactNode }) {
   }, [activeSessionIdRef, busyRef, selectedStoredSessionIdRef, updateSessionState])
 
   const { handleGatewayEvent } = useMessageStream({
+    activeGatewayProfile,
     activeSessionIdRef,
     hydrateFromStoredSession,
     queryClient,
@@ -427,7 +429,6 @@ export function ContribWiring({ children }: { children: ReactNode }) {
   // Swapping the live gateway to another profile must re-pull that profile's
   // global model + active-profile pill (both are nanostores — the blanket
   // invalidateQueries on swap doesn't touch them).
-  const activeGatewayProfile = useStore($activeGatewayProfile)
   const lastGatewayProfileRef = useRef(activeGatewayProfile)
 
   useEffect(() => {
@@ -899,12 +900,21 @@ export function ContribWiring({ children }: { children: ReactNode }) {
             void refreshCurrentModel()
             void queryClient.invalidateQueries({ queryKey: ['model-options'] })
           }}
+          profile={activeGatewayProfile}
           requestGateway={requestGateway}
         />
       )}
-      <ModelPickerOverlay gateway={gatewayRef.current || undefined} onSelect={selectModel} />
+      <ModelPickerOverlay
+        gateway={gatewayRef.current || undefined}
+        onSelect={selectModel}
+        profile={activeGatewayProfile}
+      />
       <SessionPickerOverlay onResume={resumeSession} />
-      <ModelVisibilityOverlay gateway={gatewayRef.current || undefined} onOpenProviders={openProviderSettings} />
+      <ModelVisibilityOverlay
+        gateway={gatewayRef.current || undefined}
+        onOpenProviders={openProviderSettings}
+        profile={activeGatewayProfile}
+      />
       <UpdatesOverlay />
       <GatewayConnectingOverlay />
       <BootFailureOverlay />

apps/desktop/src/app/model-picker-overlay.tsx

@@ -16,9 +16,10 @@ import { $focusedRuntimeId, $focusedSessionState } from '@/store/session-states'
 interface ModelPickerOverlayProps {
   gateway?: HermesGateway
   onSelect: (selection: ModelSelection) => void
+  profile: string
 }
 
-export function ModelPickerOverlay({ gateway, onSelect }: ModelPickerOverlayProps) {
+export function ModelPickerOverlay({ gateway, onSelect, profile }: ModelPickerOverlayProps) {
   const primarySessionId = useStore($activeSessionId)
   const primaryModel = useStore($currentModel)
   const primaryProvider = useStore($currentProvider)
@@ -45,6 +46,7 @@ export function ModelPickerOverlay({ gateway, onSelect }: ModelPickerOverlayProp
       onOpenChange={setModelPickerOpen}
       onSelect={selection => onSelect({ ...selection, sessionId })}
       open={open}
+      profile={profile}
       sessionId={sessionId}
     />
   )

apps/desktop/src/app/model-visibility-overlay.tsx

@@ -8,9 +8,10 @@ import { $activeSessionId, $gatewayState } from '@/store/session'
 interface ModelVisibilityOverlayProps {
   gateway?: HermesGateway
   onOpenProviders: () => void
+  profile: string
 }
 
-export function ModelVisibilityOverlay({ gateway, onOpenProviders }: ModelVisibilityOverlayProps) {
+export function ModelVisibilityOverlay({ gateway, onOpenProviders, profile }: ModelVisibilityOverlayProps) {
   const activeSessionId = useStore($activeSessionId)
   const gatewayOpen = useStore($gatewayState) === 'open'
   const open = useStore($modelVisibilityOpen)
@@ -25,6 +26,7 @@ export function ModelVisibilityOverlay({ gateway, onOpenProviders }: ModelVisibi
       onOpenChange={setModelVisibilityOpen}
       onOpenProviders={onOpenProviders}
       open={open}
+      profile={profile}
       sessionId={activeSessionId}
     />
   )

apps/desktop/src/app/session/hooks/use-message-stream/gateway-event.ts

@@ -12,6 +12,7 @@ import { coerceGatewayText, coerceThinkingText, normalizePersonalityValue } from
 import { playCompletionSound } from '@/lib/completion-sound'
 import { resolveGatewayEventSessionId } from '@/lib/gateway-events'
 import { triggerHaptic } from '@/lib/haptics'
+import { modelOptionsQueryKey } from '@/lib/model-options'
 import { isProviderSetupErrorMessage } from '@/lib/provider-setup-errors'
 import { reconcileApprovalModeForProfile } from '@/store/approval-mode'
 import { clearClarifyRequest, setClarifyRequest } from '@/store/clarify'
@@ -70,6 +71,7 @@ const COMPACTION_RESUME_EVENT_TYPES = new Set([
 ])
 
 interface GatewayEventDeps {
+  activeGatewayProfile: string
   activeSessionIdRef: MutableRefObject<string | null>
   compactedTurnRef: MutableRefObject<Set<string>>
   lastCwdInfoSessionRef: MutableRefObject<string | null>
@@ -102,6 +104,7 @@ export function useGatewayEventHandler(deps: GatewayEventDeps) {
   const {
     appendAssistantDelta,
     appendReasoningDelta,
+    activeGatewayProfile,
     activeSessionIdRef,
     compactedTurnRef,
     lastCwdInfoSessionRef,
@@ -372,7 +375,8 @@ export function useGatewayEventHandler(deps: GatewayEventDeps) {
 
         if (modelValueChanged || providerValueChanged) {
           void queryClient.invalidateQueries({
-            queryKey: explicitSid && sessionId ? ['model-options', sessionId] : ['model-options']
+            queryKey:
+              explicitSid && sessionId ? modelOptionsQueryKey(activeGatewayProfile, sessionId) : ['model-options']
           })
         }
       } else if (event.type === 'message.start') {
@@ -837,6 +841,7 @@ export function useGatewayEventHandler(deps: GatewayEventDeps) {
       appendAssistantDelta,
       appendReasoningDelta,
       activeSessionIdRef,
+      activeGatewayProfile,
       compactedTurnRef,
       completeAssistantMessage,
       failAssistantMessage,

apps/desktop/src/app/session/hooks/use-message-stream/index.ts

@@ -32,6 +32,7 @@ import { useGatewayEventHandler } from './gateway-event'
 import { completionErrorText, delegateTaskPayloads, STREAM_DELTA_FLUSH_MS } from './utils'
 
 interface MessageStreamOptions {
+  activeGatewayProfile?: string
   activeSessionIdRef: MutableRefObject<string | null>
   hydrateFromStoredSession: (
     attempts?: number,
@@ -55,6 +56,7 @@ interface QueuedStreamDeltas {
 }
 
 export function useMessageStream({
+  activeGatewayProfile = 'default',
   activeSessionIdRef,
   hydrateFromStoredSession,
   queryClient,
@@ -613,6 +615,7 @@ export function useMessageStream({
   )
 
   const handleGatewayEvent = useGatewayEventHandler({
+    activeGatewayProfile,
     appendAssistantDelta,
     appendReasoningDelta,
     activeSessionIdRef,

apps/desktop/src/app/session/hooks/use-message-stream/session-info-side-effects.test.tsx

@@ -5,6 +5,7 @@ import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 
 import type { ClientSessionState } from '@/app/types'
 import { createClientSessionState } from '@/lib/chat-runtime'
+import { modelOptionsQueryKey } from '@/lib/model-options'
 import { setCurrentModel, setCurrentProvider } from '@/store/session'
 import type { RpcEvent } from '@/types/hermes'
 
@@ -16,6 +17,7 @@ import { useMessageStream } from './index'
 // actually changed. message.complete must coalesce sidebar refreshes.
 
 const ACTIVE_SID = 'session-active'
+const ACTIVE_PROFILE = 'compass'
 let handleEvent: ((event: RpcEvent) => void) | null = null
 let refreshHermesConfig: ReturnType<typeof vi.fn<() => Promise<void>>>
 let refreshSessions: ReturnType<typeof vi.fn<() => Promise<void>>>
@@ -26,6 +28,7 @@ function Harness() {
   const sessionStateByRuntimeIdRef = useRef(new Map<string, ClientSessionState>())
 
   const stream = useMessageStream({
+    activeGatewayProfile: ACTIVE_PROFILE,
     activeSessionIdRef,
     hydrateFromStoredSession: vi.fn(async () => undefined),
     queryClient,
@@ -132,7 +135,7 @@ describe('session.info model-options invalidation gating', () => {
 
     sessionInfo(ACTIVE_SID, { model: 'm2', provider: 'p1', running: true })
 
-    expect(invalidate).toHaveBeenCalledWith({ queryKey: ['model-options', ACTIVE_SID] })
+    expect(invalidate).toHaveBeenCalledWith({ queryKey: modelOptionsQueryKey(ACTIVE_PROFILE, ACTIVE_SID) })
   })
 })
 

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

@@ -3,6 +3,8 @@ import { cleanup, render, renderHook } from '@testing-library/react'
 import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 
 import { getGlobalModelInfo } from '@/hermes'
+import { modelOptionsQueryKey } from '@/lib/model-options'
+import { $activeGatewayProfile } from '@/store/profile'
 import {
   $activeSessionId,
   $currentModel,
@@ -79,6 +81,7 @@ function Harness({
 
 describe('useModelControls', () => {
   beforeEach(() => {
+    $activeGatewayProfile.set('default')
     $activeSessionId.set(null)
     setCurrentModel('')
     setCurrentModelSource('')
@@ -88,6 +91,7 @@ describe('useModelControls', () => {
   afterEach(() => {
     cleanup()
     vi.restoreAllMocks()
+    $activeGatewayProfile.set('default')
     $activeSessionId.set(null)
     setCurrentModel('')
     setCurrentModelSource('')
@@ -201,6 +205,26 @@ describe('useModelControls', () => {
     expect(setGlobalModel).not.toHaveBeenCalled()
   })
 
+  it('updates only the active profile new-chat cache', async () => {
+    const queryClient = new QueryClient()
+    $activeGatewayProfile.set('compass')
+
+    const { result } = renderHook(() =>
+      useModelControls({
+        queryClient,
+        requestGateway: vi.fn()
+      })
+    )
+
+    await result.current.selectModel({ model: 'qwen3.6:35b-65k', provider: 'custom:local-ollama' })
+
+    expect(queryClient.getQueryData(modelOptionsQueryKey('compass'))).toMatchObject({
+      model: 'qwen3.6:35b-65k',
+      provider: 'custom:local-ollama'
+    })
+    expect(queryClient.getQueryData(modelOptionsQueryKey('default'))).toBeUndefined()
+  })
+
   it('seeds an empty composer model from global but never clobbers a pick', async () => {
     vi.mocked(getGlobalModelInfo).mockResolvedValue({ model: 'openai/gpt-5.5', provider: 'openai-codex' })
 
@@ -232,7 +256,11 @@ describe('useModelControls', () => {
     vi.mocked(getGlobalModelInfo).mockResolvedValue({ model: 'openai/gpt-5.5', provider: 'openai-codex' })
 
     const queryClient = new QueryClient()
-    queryClient.setQueryData(['model-options', 'global'], {
+    $activeGatewayProfile.set('compass')
+    queryClient.setQueryData(modelOptionsQueryKey('default'), {
+      providers: [{ models: ['openrouter/owl-alpha'], name: 'OpenRouter', slug: 'openrouter' }]
+    })
+    queryClient.setQueryData(modelOptionsQueryKey('compass'), {
       providers: [{ models: ['openai/gpt-5.5'], name: 'OpenRouter', slug: 'openrouter' }]
     })
 
@@ -253,7 +281,7 @@ describe('useModelControls', () => {
     vi.mocked(getGlobalModelInfo).mockResolvedValue({ model: 'openai/gpt-5.5', provider: 'openai-codex' })
 
     const queryClient = new QueryClient()
-    queryClient.setQueryData(['model-options', 'global'], {
+    queryClient.setQueryData(modelOptionsQueryKey('default'), {
       providers: [{ models: ['openrouter/glm-4.7', 'openai/gpt-5.5'], name: 'OpenRouter', slug: 'openrouter' }]
     })
 
@@ -346,16 +374,16 @@ describe('useModelControls', () => {
   })
 
   it('targets an explicit tile sessionId without clobbering the primary model', async () => {
+    const queryClient = new QueryClient()
+    $activeGatewayProfile.set('compass')
     $activeSessionId.set('primary-runtime')
     setCurrentModel('primary/model')
     setCurrentProvider('openai')
     const requestGateway = vi.fn(async () => ({ key: 'model', value: 'tile-model' }) as never)
-    let controls!: Controls
-
-    render(<Harness onReady={value => (controls = value)} requestGateway={requestGateway} />)
+    const { result } = renderHook(() => useModelControls({ queryClient, requestGateway }))
 
     await expect(
-      controls.selectModel({
+      result.current.selectModel({
         model: 'tile-model',
         provider: 'anthropic',
         sessionId: 'tile-runtime'
@@ -370,5 +398,10 @@ describe('useModelControls', () => {
     // Primary footer untouched — the busy primary must not absorb a tile pick.
     expect($currentModel.get()).toBe('primary/model')
     expect($currentProvider.get()).toBe('openai')
+    expect(queryClient.getQueryData(modelOptionsQueryKey('compass', 'tile-runtime'))).toMatchObject({
+      model: 'tile-model',
+      provider: 'anthropic'
+    })
+    expect(queryClient.getQueryData(modelOptionsQueryKey('default', 'tile-runtime'))).toBeUndefined()
   })
 })

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

@@ -4,8 +4,9 @@ import { useCallback, useRef } from 'react'
 import type { ModelSelection } from '@/app/shell/model-menu-panel'
 import { getGlobalModelInfo } from '@/hermes'
 import { useI18n } from '@/i18n'
-import { manualPickRemoved } from '@/lib/model-options'
+import { manualPickRemoved, modelOptionsQueryKey } from '@/lib/model-options'
 import { notifyError } from '@/store/notifications'
+import { $activeGatewayProfile } from '@/store/profile'
 import {
   $activeSessionId,
   $currentModel,
@@ -36,13 +37,19 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
   // callbacks once and never re-evaluate — a captured prop would be stale
   // forever. The store read is always current.
   const updateModelOptionsCache = useCallback(
-    (sessionId: null | string, provider: string, model: string, includeGlobal: boolean) => {
+    (
+      sessionId: null | string,
+      provider: string,
+      model: string,
+      includeGlobal: boolean,
+      profile = $activeGatewayProfile.get()
+    ) => {
       const patch = (prev: ModelOptionsResponse | undefined) => ({ ...(prev ?? {}), provider, model })
 
-      queryClient.setQueryData<ModelOptionsResponse>(['model-options', sessionId || 'global'], patch)
+      queryClient.setQueryData<ModelOptionsResponse>(modelOptionsQueryKey(profile, sessionId), patch)
 
       if (includeGlobal) {
-        queryClient.setQueryData<ModelOptionsResponse>(['model-options', 'global'], patch)
+        queryClient.setQueryData<ModelOptionsResponse>(modelOptionsQueryKey(profile), patch)
       }
     },
     [queryClient]
@@ -78,7 +85,9 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
             return false
           }
 
-          const options = queryClient.getQueryData<ModelOptionsResponse>(['model-options', 'global'])
+          const options = queryClient.getQueryData<ModelOptionsResponse>(
+            modelOptionsQueryKey($activeGatewayProfile.get())
+          )
 
           return !manualPickRemoved(options?.providers, $currentProvider.get(), $currentModel.get())
         }
@@ -144,6 +153,7 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
         : ($sessionStates.get()[liveSessionId!]?.provider ?? '')
 
       const prevSource = getCurrentModelSource()
+      const liveGatewayProfile = $activeGatewayProfile.get()
 
       if (touchesPrimary) {
         setCurrentModel(selection.model)
@@ -158,7 +168,13 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
         }))
       }
 
-      updateModelOptionsCache(liveSessionId, selection.provider, selection.model, touchesPrimary && !liveSessionId)
+      updateModelOptionsCache(
+        liveSessionId,
+        selection.provider,
+        selection.model,
+        touchesPrimary && !liveSessionId,
+        liveGatewayProfile
+      )
 
       // No live session yet: the pick is pure UI state. session.create reads
       // $currentModel/$currentProvider and applies it as that session's override.
@@ -173,7 +189,7 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
           value: `${selection.model} --provider ${selection.provider} --session`
         })
 
-        void queryClient.invalidateQueries({ queryKey: ['model-options', liveSessionId] })
+        void queryClient.invalidateQueries({ queryKey: modelOptionsQueryKey(liveGatewayProfile, liveSessionId) })
 
         return true
       } catch (err) {
@@ -189,7 +205,13 @@ export function useModelControls({ queryClient, requestGateway }: ModelControlsO
           }))
         }
 
-        updateModelOptionsCache(liveSessionId, prevProvider, prevModel, touchesPrimary && !liveSessionId)
+        updateModelOptionsCache(
+          liveSessionId,
+          prevProvider,
+          prevModel,
+          touchesPrimary && !liveSessionId,
+          liveGatewayProfile
+        )
         notifyError(err, copy.modelSwitchFailed)
 
         return false

apps/desktop/src/app/shell/model-menu-panel.tsx

@@ -19,7 +19,7 @@ import { Skeleton } from '@/components/ui/skeleton'
 import type { HermesGateway } from '@/hermes'
 import { useI18n } from '@/i18n'
 import { ChevronDown, ChevronRight } from '@/lib/icons'
-import { requestModelOptions } from '@/lib/model-options'
+import { modelOptionsQueryKey, requestModelOptions } from '@/lib/model-options'
 import {
   currentPickerSelection,
   displayModelName,
@@ -59,6 +59,7 @@ export interface ModelSelection {
 interface ModelMenuPanelProps {
   gateway?: HermesGateway
   onSelectModel: (selection: ModelSelection) => Promise<boolean> | void
+  profile?: string
   requestGateway: <T>(method: string, params?: Record<string, unknown>) => Promise<T>
 }
 
@@ -67,7 +68,7 @@ interface ProviderGroup {
   provider: ModelOptionProvider
 }
 
-export function ModelMenuPanel({ gateway, onSelectModel, requestGateway }: ModelMenuPanelProps) {
+export function ModelMenuPanel({ gateway, onSelectModel, profile = 'default', requestGateway }: ModelMenuPanelProps) {
   const { t } = useI18n()
   const copy = t.shell.modelMenu
   const closeMenu = useContext(ModelMenuCloseContext)
@@ -87,7 +88,7 @@ export function ModelMenuPanel({ gateway, onSelectModel, requestGateway }: Model
   const collapsedProviders = useStore($collapsedProviders)
 
   const modelOptions = useQuery({
-    queryKey: ['model-options', activeSessionId || 'global'],
+    queryKey: modelOptionsQueryKey(profile, activeSessionId),
     // Gateway-first even with no session yet: a connected (possibly remote)
     // gateway owns the model catalog, including virtual providers like `moa`
     // that the local REST fallback can't know about (#53817).
@@ -148,7 +149,7 @@ export function ModelMenuPanel({ gateway, onSelectModel, requestGateway }: Model
     setRefreshing(true)
 
     try {
-      const queryKey = ['model-options', activeSessionId || 'global']
+      const queryKey = modelOptionsQueryKey(profile, activeSessionId)
 
       const next = await requestModelOptions({ gateway, refresh: true, sessionId: activeSessionId })
 

apps/desktop/src/components/model-picker.tsx

@@ -2,7 +2,7 @@ import { useQuery } from '@tanstack/react-query'
 import { useState } from 'react'
 
 import { useI18n } from '@/i18n'
-import { requestModelOptions } from '@/lib/model-options'
+import { modelOptionsQueryKey, requestModelOptions } from '@/lib/model-options'
 import { currentPickerSelection } from '@/lib/model-status-label'
 import { normalize } from '@/lib/text'
 import type { ModelOptionProvider, ModelPricing } from '@/types/hermes'
@@ -25,6 +25,7 @@ interface ModelPickerDialogProps {
   currentModel: string
   currentProvider: string
   onSelect: (selection: { provider: string; model: string }) => void
+  profile?: string
   /**
    * Optional class to apply to DialogContent. Use to override z-index when
    * stacking the picker on top of another fixed overlay (e.g. the desktop
@@ -42,6 +43,7 @@ export function ModelPickerDialog({
   currentModel,
   currentProvider,
   onSelect,
+  profile = 'default',
   contentClassName
 }: ModelPickerDialogProps) {
   const { t } = useI18n()
@@ -54,7 +56,7 @@ export function ModelPickerDialog({
   const [search, setSearch] = useState('')
 
   const modelOptions = useQuery({
-    queryKey: ['model-options', sessionId || 'global'],
+    queryKey: modelOptionsQueryKey(profile, sessionId),
     queryFn: () => requestModelOptions({ gateway: gw, sessionId }),
     enabled: open
   })

apps/desktop/src/components/model-visibility-dialog.tsx

@@ -7,8 +7,8 @@ import { Dialog, DialogContent, DialogHeader, DialogTitle } from '@/components/u
 import { GlyphSpinner } from '@/components/ui/glyph-spinner'
 import { Switch } from '@/components/ui/switch'
 import type { HermesGateway } from '@/hermes'
-import { getGlobalModelOptions } from '@/hermes'
 import { useI18n } from '@/i18n'
+import { modelOptionsQueryKey, requestModelOptions } from '@/lib/model-options'
 import { displayModelName, modelDisplayParts } from '@/lib/model-status-label'
 import { normalize } from '@/lib/text'
 import {
@@ -26,6 +26,7 @@ interface ModelVisibilityDialogProps {
   onOpenChange: (open: boolean) => void
   onOpenProviders: () => void
   open: boolean
+  profile?: string
   sessionId?: string | null
 }
 
@@ -34,6 +35,7 @@ export function ModelVisibilityDialog({
   onOpenChange,
   onOpenProviders,
   open,
+  profile = 'default',
   sessionId
 }: ModelVisibilityDialogProps) {
   const { t } = useI18n()
@@ -42,17 +44,8 @@ export function ModelVisibilityDialog({
   const stored = useStore($visibleModels)
 
   const modelOptions = useQuery({
-    queryKey: ['model-options', sessionId || 'global'],
-    queryFn: (): Promise<ModelOptionsResponse> => {
-      if (gw && sessionId) {
-        return gw.request<ModelOptionsResponse>('model.options', {
-          session_id: sessionId,
-          explicit_only: true
-        })
-      }
-
-      return getGlobalModelOptions()
-    },
+    queryKey: modelOptionsQueryKey(profile, sessionId),
+    queryFn: (): Promise<ModelOptionsResponse> => requestModelOptions({ gateway: gw, sessionId }),
     enabled: open
   })
 

apps/desktop/src/components/onboarding/flow.tsx

@@ -52,7 +52,7 @@ export function FlowPanel({
   }
 
   if (flow.status === 'confirming_model') {
-    return <ConfirmingModelPanel flow={flow} leaving={leaving} onBegin={onBegin} />
+    return <ConfirmingModelPanel flow={flow} leaving={leaving} onBegin={onBegin} profile={ctx.profile} />
   }
 
   if (flow.status === 'error') {
@@ -217,11 +217,13 @@ function CancelBtn({ size = 'default' }: { size?: 'default' | 'sm' }) {
 function ConfirmingModelPanel({
   flow,
   leaving,
-  onBegin
+  onBegin,
+  profile
 }: {
   flow: Extract<OnboardingFlow, { status: 'confirming_model' }>
   leaving: boolean
   onBegin: () => void
+  profile?: string
 }) {
   const { t } = useI18n()
   const scrambledModel = useScramble(flow.currentModel, leaving)
@@ -323,6 +325,7 @@ function ConfirmingModelPanel({
           setPickerOpen(false)
         }}
         open={pickerOpen}
+        profile={profile}
       />
     </div>
   )

apps/desktop/src/components/onboarding/index.tsx

@@ -49,6 +49,7 @@ export {
 interface DesktopOnboardingOverlayProps {
   enabled: boolean
   onCompleted?: () => void
+  profile: string
   requestGateway: OnboardingContext['requestGateway']
 }
 
@@ -177,17 +178,25 @@ function useApiKeyCatalog(): ApiKeyOption[] {
 // → surface-out (520ms, held back by [transition-delay:660ms]). Finalize after.
 const ONBOARDING_EXIT_MS = 1180
 
-export function DesktopOnboardingOverlay({ enabled, onCompleted, requestGateway }: DesktopOnboardingOverlayProps) {
+export function DesktopOnboardingOverlay({
+  enabled,
+  onCompleted,
+  profile,
+  requestGateway
+}: DesktopOnboardingOverlayProps) {
   const { t } = useI18n()
   const onboarding = useStore($desktopOnboarding)
   const boot = useStore($desktopBoot)
-  const ctxRef = useRef<OnboardingContext>({ requestGateway, onCompleted })
-  ctxRef.current = { requestGateway, onCompleted }
+  const ctxRef = useRef<OnboardingContext>({ requestGateway, onCompleted, profile })
+  ctxRef.current = { requestGateway, onCompleted, profile }
 
   const ctx = useMemo<OnboardingContext>(
     () => ({
       requestGateway: (...args) => ctxRef.current.requestGateway(...args),
-      onCompleted: () => ctxRef.current.onCompleted?.()
+      onCompleted: () => ctxRef.current.onCompleted?.(),
+      get profile() {
+        return ctxRef.current.profile
+      }
     }),
     []
   )

apps/desktop/src/lib/model-options.test.ts

@@ -2,7 +2,7 @@ import { afterEach, describe, expect, it, vi } from 'vitest'
 
 import { getGlobalModelOptions } from '@/hermes'
 
-import { manualPickRemoved, requestModelOptions } from './model-options'
+import { manualPickRemoved, modelOptionsQueryKey, requestModelOptions } from './model-options'
 
 const globalOptions = { model: 'hermes-4', provider: 'nous', providers: [] }
 
@@ -49,6 +49,18 @@ describe('requestModelOptions', () => {
   })
 })
 
+describe('modelOptionsQueryKey', () => {
+  it('isolates new-chat catalogs by active gateway profile', () => {
+    expect(modelOptionsQueryKey('default')).toEqual(['model-options', 'default', 'global'])
+    expect(modelOptionsQueryKey('compass')).toEqual(['model-options', 'compass', 'global'])
+    expect(modelOptionsQueryKey('default')).not.toEqual(modelOptionsQueryKey('compass'))
+  })
+
+  it('keeps session catalogs inside the owning profile namespace', () => {
+    expect(modelOptionsQueryKey(' compass ', 'session-1')).toEqual(['model-options', 'compass', 'session-1'])
+  })
+})
+
 describe('manualPickRemoved', () => {
   const providers = [
     { name: 'OpenRouter', slug: 'openrouter', models: ['owl-alpha', 'gpt-5.5'] },

apps/desktop/src/lib/model-options.ts

@@ -44,6 +44,12 @@ interface ModelOptionsRequest {
   sessionId?: null | string
 }
 
+export function modelOptionsQueryKey(profile: null | string | undefined, sessionId?: null | string) {
+  const profileKey = (profile ?? '').trim() || 'default'
+
+  return ['model-options', profileKey, sessionId || 'global'] as const
+}
+
 export function requestModelOptions({
   explicitOnly = true,
   gateway,

apps/desktop/src/store/onboarding.ts

@@ -76,6 +76,7 @@ export interface DesktopOnboardingState {
 
 export interface OnboardingContext {
   onCompleted?: () => void
+  profile?: string
   requestGateway: <T = unknown>(method: string, params?: Record<string, unknown>) => Promise<T>
 }
 

tests/test_tui_gateway_server.py

@@ -7,7 +7,7 @@ import time
 import types
 from datetime import datetime
 from pathlib import Path
-from unittest.mock import patch
+from unittest.mock import Mock, patch
 
 import pytest
 
@@ -8156,6 +8156,132 @@ def test_model_options_hides_unconfigured_providers_by_default(monkeypatch):
     assert calls[-1]["include_unconfigured"] is True
 
 
+def test_model_options_preserves_canonical_custom_row_after_agent_init(monkeypatch):
+    from hermes_cli.inventory import ConfigContext
+
+    class _Agent:
+        provider = "custom"
+        model = "qwen3.6:35b-65k"
+        base_url = "http://127.0.0.1:11434/v1"
+
+    server._sessions["custom-session"] = _session(agent=_Agent())
+    monkeypatch.setattr(server, "_resolve_model", lambda: "")
+    monkeypatch.setattr(
+        "hermes_cli.inventory.load_picker_context",
+        lambda: ConfigContext(
+            current_provider="custom:local-ollama",
+            current_model="qwen3.6:35b-65k",
+            current_base_url="http://127.0.0.1:11434/v1",
+            user_providers={},
+            custom_providers=[],
+        ),
+    )
+    canonical = Mock(return_value="custom:local-ollama")
+    monkeypatch.setattr(
+        "hermes_cli.runtime_provider.canonical_custom_identity",
+        canonical,
+    )
+    monkeypatch.setattr(
+        "hermes_cli.model_switch.list_authenticated_providers",
+        lambda **_kwargs: [
+            {
+                "slug": "custom:local-ollama",
+                "name": "Local Ollama",
+                "is_current": True,
+                "is_user_defined": True,
+                "models": ["qwen3.6:35b-65k"],
+                "total_models": 1,
+            },
+            {
+                "slug": "anthropic",
+                "name": "Anthropic",
+                "is_current": False,
+                "is_user_defined": False,
+                "models": ["claude-sonnet-4.6"],
+                "total_models": 1,
+            },
+        ],
+    )
+    monkeypatch.setattr(
+        "hermes_cli.auth.is_provider_explicitly_configured",
+        lambda _slug: False,
+    )
+    monkeypatch.setattr("hermes_cli.inventory._apply_pricing", lambda *_args, **_kwargs: None)
+    monkeypatch.setattr("hermes_cli.inventory._apply_capabilities", lambda *_args, **_kwargs: None)
+
+    resp = server._methods["model.options"](
+        102,
+        {"session_id": "custom-session", "explicit_only": True},
+    )
+
+    assert "result" in resp, resp
+    assert resp["result"]["provider"] == "custom:local-ollama"
+    assert [row["slug"] for row in resp["result"]["providers"]] == [
+        "custom:local-ollama"
+    ]
+    canonical.assert_called_once_with(
+        base_url="http://127.0.0.1:11434/v1",
+        config_provider="custom:local-ollama",
+    )
+
+
+def test_model_save_key_uses_credential_lifecycle_and_picker_context(monkeypatch):
+    env_var = "TEST_PROVIDER_API_KEY"
+    agent = object()
+    picker_ctx = object()
+    provider = {
+        "slug": "test-provider",
+        "name": "Test Provider",
+        "models": ["test-model"],
+        "total_models": 1,
+    }
+    server._sessions["save-key-session"] = _session(agent=agent)
+    monkeypatch.setattr(
+        "hermes_cli.auth.PROVIDER_REGISTRY",
+        {
+            "test-provider": types.SimpleNamespace(
+                name="Test Provider",
+                auth_type="api_key",
+                api_key_env_vars=(env_var,),
+            )
+        },
+    )
+    monkeypatch.setattr("hermes_cli.config.is_managed", lambda: False)
+    save_credential = Mock()
+    monkeypatch.setattr(
+        "hermes_cli.credential_lifecycle.save_provider_env_credential",
+        save_credential,
+    )
+    picker_context = Mock(return_value=picker_ctx)
+    monkeypatch.setattr(server, "_model_picker_context", picker_context)
+    build_payload = Mock(return_value={"providers": [provider]})
+    monkeypatch.setattr(
+        "hermes_cli.inventory.build_models_payload",
+        build_payload,
+    )
+    monkeypatch.setenv(env_var, "previous-value")
+    fake_key = "replacement-" + "value"
+
+    resp = server._methods["model.save_key"](
+        103,
+        {
+            "slug": "test-provider",
+            "api_key": fake_key,
+            "session_id": "save-key-session",
+        },
+    )
+
+    assert "result" in resp, resp
+    assert resp["result"]["provider"] == {**provider, "authenticated": True}
+    save_credential.assert_called_once_with(env_var, fake_key)
+    picker_context.assert_called_once_with(agent)
+    build_payload.assert_called_once_with(
+        picker_ctx,
+        picker_hints=True,
+        max_models=50,
+    )
+
+
 # ---------------------------------------------------------------------------
 # prompt.submit — auto-title
 # ---------------------------------------------------------------------------

tui_gateway/server.py

@@ -14542,10 +14542,42 @@ def _(rid, params: dict) -> dict:
         return _err(rid, 5020, str(e))
 
 
+def _model_picker_context(agent):
+    """Layer live session state onto config without losing custom identity."""
+    from hermes_cli.inventory import load_picker_context
+
+    ctx = load_picker_context()
+    provider = getattr(agent, "provider", "") if agent else ""
+    base_url = getattr(agent, "base_url", "") if agent else ""
+    if str(provider or "").strip().lower() == "custom":
+        try:
+            from hermes_cli.runtime_provider import canonical_custom_identity
+
+            provider = (
+                canonical_custom_identity(
+                    base_url=base_url or None,
+                    config_provider=ctx.current_provider,
+                )
+                or provider
+            )
+        except Exception:
+            logger.debug(
+                "custom provider identity recovery failed (model picker)",
+                exc_info=True,
+            )
+
+    return ctx.with_overrides(
+        current_provider=provider,
+        current_model=(getattr(agent, "model", "") if agent else "")
+        or _resolve_model(),
+        current_base_url=base_url,
+    )
+
+
 @method("model.options")
 def _(rid, params: dict) -> dict:
     try:
-        from hermes_cli.inventory import build_models_payload, load_picker_context
+        from hermes_cli.inventory import build_models_payload
 
         session = _sessions.get(params.get("session_id", ""))
         agent = session.get("agent") if session else None
@@ -14553,13 +14585,7 @@ def _(rid, params: dict) -> dict:
         # is spawned, IT owns the live provider/model/base_url. Empty
         # agent attributes must NOT clobber disk config (with_overrides
         # is truthy-only).
-        ctx = load_picker_context().with_overrides(
-            current_provider=getattr(agent, "provider", "") if agent else "",
-            current_model=(
-                (getattr(agent, "model", "") if agent else "") or _resolve_model()
-            ),
-            current_base_url=getattr(agent, "base_url", "") if agent else "",
-        )
+        ctx = _model_picker_context(agent)
         # picker_hints + canonical_order produce the TUI/desktop picker shape:
         # `authenticated`/`auth_type`/`key_env`/`warning` per row, in
         # CANONICAL_PROVIDERS declaration order. Desktop pickers default to the
@@ -14599,7 +14625,7 @@ def _(rid, params: dict) -> dict:
     try:
         from hermes_cli.auth import PROVIDER_REGISTRY
         from hermes_cli.config import is_managed
-        from hermes_cli.inventory import build_models_payload, load_picker_context
+        from hermes_cli.inventory import build_models_payload
 
         slug = (params.get("slug") or "").strip()
         api_key = (params.get("api_key") or "").strip()
@@ -14640,13 +14666,7 @@ def _(rid, params: dict) -> dict:
         # carries `authenticated` for the TUI frontend.
         session = _sessions.get(params.get("session_id", ""))
         agent = session.get("agent") if session else None
-        ctx = load_picker_context().with_overrides(
-            current_provider=getattr(agent, "provider", "") if agent else "",
-            current_model=(
-                (getattr(agent, "model", "") if agent else "") or _resolve_model()
-            ),
-            current_base_url=getattr(agent, "base_url", "") if agent else "",
-        )
+        ctx = _model_picker_context(agent)
         payload = build_models_payload(
             ctx, picker_hints=True, max_models=50,
         )