feat(desktop): auto-speak text responses via voice.auto_tts config

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

@@ -25,6 +25,7 @@ import { desktopSlashCommandTakesArgs } from '@/lib/desktop-slash-commands'
 import { DATA_IMAGE_URL_RE } from '@/lib/embedded-images'
 import { triggerHaptic } from '@/lib/haptics'
 import { cn } from '@/lib/utils'
+import { isVoicePlaybackActive, playSpeechText } from '@/lib/voice-playback'
 import {
   $composerAttachments,
   clearComposerAttachments,
@@ -157,6 +158,7 @@ const cloneAttachments = (attachments: ComposerAttachment[]) => attachments.map(
 const DRAFT_PERSIST_DEBOUNCE_MS = 400
 
 export function ChatBar({
+  autoTtsEnabled,
   busy,
   cwd,
   disabled,
@@ -1528,6 +1530,38 @@ export function ChatBar({
     }
   }, [autoDrainNext, busy, queuedPrompts.length])
 
+  // Auto-TTS: speak last assistant message on busy → false, when voice.auto_tts is on.
+  const wasBusyRef = useRef(false)
+  const autoTtsSessionRef = useRef(sessionId)
+
+  // Reset on session switch — ChatBar is persistent, refs survive across sessions.
+  if (autoTtsSessionRef.current !== sessionId) {
+    autoTtsSessionRef.current = sessionId
+    wasBusyRef.current = false
+  }
+
+  useEffect(() => {
+    if (busy) {
+      wasBusyRef.current = true
+
+      return
+    }
+
+    if (!wasBusyRef.current || !autoTtsEnabled || voiceConversationActive || isVoicePlaybackActive()) {
+      return
+    }
+
+    const response = pendingResponse()
+
+    if (!response || response.pending) {
+      return
+    }
+
+    // Consume before async call to prevent double-play on a second effect fire.
+    consumePendingResponse()
+    void playSpeechText(response.text, { messageId: response.id, source: 'read-aloud' })
+  }, [busy]) // eslint-disable-line react-hooks/exhaustive-deps
+
   // Queue-edit cleanup: on session swap the scope effect already stashed the
   // edit snapshot; only restore into the composer when still on the same scope.
   useEffect(() => {
@@ -1899,6 +1933,7 @@ export function ChatBar({
                   composerSurfaceGlass
                 )}
               />
+              <VoicePlaybackActivity />
               <div
                 className={cn(
                   'relative z-1 flex min-h-0 w-full flex-col gap-(--composer-row-gap) overflow-hidden rounded-[inherit] px-(--composer-surface-pad-x) py-(--composer-surface-pad-y) transition-opacity duration-200 ease-out',
@@ -1907,7 +1942,6 @@ export function ChatBar({
                 data-slot="composer-fade"
               >
                 <VoiceActivity state={voiceActivityState} />
-                <VoicePlaybackActivity />
                 {queueEdit && editingQueuedPrompt && (
                   <div className="flex items-center justify-between gap-2 rounded-lg border border-[color-mix(in_srgb,var(--dt-composer-ring)_32%,transparent)] bg-accent/18 px-2 py-1">
                     <div className="min-w-0 text-[0.7rem] text-muted-foreground/88">

apps/desktop/src/app/chat/composer/types.ts

@@ -28,6 +28,7 @@ export interface ChatBarState {
 }
 
 export interface ChatBarProps {
+  autoTtsEnabled?: boolean
   busy: boolean
   disabled: boolean
   focusKey?: string | null

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

@@ -60,6 +60,7 @@ import { SessionActionsMenu } from './sidebar/session-actions-menu'
 import { threadLoadingState } from './thread-loading'
 
 interface ChatViewProps extends Omit<React.ComponentProps<'div'>, 'onSubmit'> {
+  autoTtsEnabled?: boolean
   gateway: HermesGateway | null
   onToggleSelectedPin: () => void
   onDeleteSelectedSession: () => void
@@ -256,6 +257,7 @@ export function ChatView({
   onAddUrl,
   onAttachImageBlob,
   onAttachDroppedItems,
+  autoTtsEnabled,
   onBranchInNewChat,
   maxVoiceRecordingSeconds,
   onPasteClipboardImage,
@@ -432,6 +434,7 @@ export function ChatView({
           {showChatBar && (
             <Suspense fallback={<ChatBarFallback />}>
               <ChatBar
+                autoTtsEnabled={autoTtsEnabled}
                 busy={busy}
                 cwd={currentCwd}
                 disabled={!gatewayOpen}

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

@@ -530,7 +530,7 @@ export function DesktopController() {
     requestGateway
   })
 
-  const { refreshHermesConfig, sttEnabled, voiceMaxRecordingSeconds } = useHermesConfig({
+  const { autoTtsEnabled, refreshHermesConfig, sttEnabled, voiceMaxRecordingSeconds } = useHermesConfig({
     activeSessionIdRef,
     refreshProjectBranch
   })
@@ -967,6 +967,7 @@ export function DesktopController() {
 
   const chatView = (
     <ChatView
+      autoTtsEnabled={autoTtsEnabled}
       gateway={gatewayRef.current}
       maxVoiceRecordingSeconds={voiceMaxRecordingSeconds}
       onAddContextRef={composer.addContextRefAttachment}

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

@@ -28,6 +28,7 @@ interface HermesConfigOptions {
 export function useHermesConfig({ activeSessionIdRef, refreshProjectBranch }: HermesConfigOptions) {
   const [voiceMaxRecordingSeconds, setVoiceMaxRecordingSeconds] = useState(DEFAULT_VOICE_SECONDS)
   const [sttEnabled, setSttEnabled] = useState(true)
+  const [autoTtsEnabled, setAutoTtsEnabled] = useState(false)
 
   const refreshHermesConfig = useCallback(async () => {
     try {
@@ -65,10 +66,11 @@ export function useHermesConfig({ activeSessionIdRef, refreshProjectBranch }: He
 
       setVoiceMaxRecordingSeconds(recordingLimit(config.voice?.max_recording_seconds))
       setSttEnabled(config.stt?.enabled !== false)
+      setAutoTtsEnabled(config.voice?.auto_tts === true)
     } catch {
       // Config is nice-to-have; chat still works without it.
     }
   }, [activeSessionIdRef, refreshProjectBranch])
 
-  return { refreshHermesConfig, sttEnabled, voiceMaxRecordingSeconds }
+  return { autoTtsEnabled, refreshHermesConfig, sttEnabled, voiceMaxRecordingSeconds }
 }

apps/desktop/src/hermes.ts

@@ -741,7 +741,10 @@ export function speakText(text: string): Promise<AudioSpeakResponse> {
   return window.hermesDesktop.api<AudioSpeakResponse>({
     path: '/api/audio/speak',
     method: 'POST',
-    body: { text }
+    body: { text },
+    // Long-response synthesis exceeds the app-wide 15s default; the backend
+    // caps edge-tts at 60s (tools/tts_tool.py), so give the HTTP layer headroom
+    timeoutMs: 75_000
   })
 }
 

apps/desktop/src/types/hermes.ts

@@ -182,6 +182,7 @@ export interface HermesConfig {
     enabled?: boolean
   }
   voice?: {
+    auto_tts?: boolean
     max_recording_seconds?: number
   }
 }

tui_gateway/server.py

@@ -6032,14 +6032,18 @@ def _stream(delta):
                     pass
 
             # CLI parity: when voice-mode TTS is on, speak the agent reply
-            # (cli.py:_voice_speak_response).  Only the final text — tool
-            # calls / reasoning already stream separately and would be
-            # noisy to read aloud.
+            # WS clients (desktop, dashboard) do their own client-side playback,
+            # so speaking here too would double the audio — restrict server-side
+            # speech to non-WS transports (TUI / stdio CLI). Class-name check
+            # avoids a circular import of tui_gateway.ws.
+            _transport = session.get("transport")
+            _is_ws_client = type(_transport).__name__ == "WSTransport"
             if (
                 status == "complete"
                 and isinstance(raw, str)
                 and raw.strip()
                 and _voice_tts_enabled()
+                and not _is_ws_client
             ):
                 try:
                     from hermes_cli.voice import speak_text
@@ -9230,8 +9234,20 @@ def _voice_mode_enabled() -> bool:
 
 
 def _voice_tts_enabled() -> bool:
-    """Whether agent replies should be spoken back via TTS (runtime only)."""
-    return os.environ.get("HERMES_VOICE_TTS", "").strip() == "1"
+    """Whether agent replies should be spoken back via TTS.
+
+    Runtime flag (``HERMES_VOICE_TTS`` env var) takes precedence so an
+    in-session toggle survives config reloads.  When the flag has never been
+    set this session, fall back to ``voice.auto_tts`` in config.yaml so the
+    setting takes effect immediately on first launch without a manual toggle.
+    """
+    raw = os.environ.get("HERMES_VOICE_TTS", "")
+    if raw.strip():
+        return raw.strip() == "1"
+    # Seed from config on first access (flag not yet set this session).
+    enabled = bool(_voice_cfg_dict().get("auto_tts", False))
+    os.environ["HERMES_VOICE_TTS"] = "1" if enabled else "0"
+    return enabled
 
 
 def _voice_cfg_dict() -> dict:
@@ -9334,8 +9350,6 @@ def _(rid, params: dict) -> dict:
         )
 
     if action == "tts":
-        if not _voice_mode_enabled():
-            return _err(rid, 4014, "enable voice mode first: /voice on")
         new_value = not _voice_tts_enabled()
         # Runtime-only flag (CLI parity) — see voice.toggle on/off above.
         os.environ["HERMES_VOICE_TTS"] = "1" if new_value else "0"