@alia.onl/sdk 4.1.0 → 5.0.0

package/CHANGELOG.md

@@ -1,5 +1,53 @@
 # @alia.onl/sdk
 
+## 5.0.0
+
+### Breaking
+
+- **Voice is now an injected capability on `AliaChatScreen` and
+  `AliaChatSheet`.** Pass `voiceSession={VoiceSession}` (from
+  `@alia.onl/sdk/voice`) to offer voice calls; without it the chat is text-only
+  and the microphone button is not rendered.
+
+  ```tsx
+  import { AliaChatScreen } from '@alia.onl/sdk';
+  import { VoiceSession } from '@alia.onl/sdk/voice';
+
+  <AliaChatScreen voiceSession={VoiceSession} />;
+  ```
+
+  `voiceSession` may be a `React.lazy` component — the chat renders it inside
+  its own `Suspense` boundary.
+
+  v4.0.0 moved the voice exports out of the root barrel so that text-chat
+  consumers would stop compiling `livekit-client`, but `AliaChatContent` still
+  imported `useVoiceRoom` and `useAudioLevelMonitor` directly from their
+  modules, and `AliaChatScreen`/`AliaChatSheet` reach it. Since this package
+  ships raw source, that put the whole LiveKit client (1,204,825 bytes raw,
+  ~250 KiB gzip) back into every text-chat consumer's module graph. The split
+  could not be made real while the shell owned voice, so it no longer does.
+
+### Added
+
+- `VoiceSession` (`@alia.onl/sdk/voice`) — the LiveKit half of the chat as one
+  component. It dials on mount, ends the call on unmount, renders
+  `VoiceControls`, and reports room state, agent state, audio amplitude, and the
+  transcript back to the chat shell.
+- `VoiceSessionComponent`, `VoiceSessionProps`, and `VoiceSessionState` types,
+  exported from both entries.
+
+### Fixed
+
+- An unexpected room drop now always ends voice mode. It previously required a
+  non-empty pre-call transcript, so a dropped call in a fresh chat left the
+  voice UI up with no room behind it.
+
+### Internal
+
+- `bun run --filter @alia.onl/sdk check:entries` walks the real import graph and
+  fails if the root entry can reach `livekit-client` (or if the voice entry
+  cannot). Wired into CI alongside a new SDK typecheck step.
+
 ## 4.1.0
 
 ### Changed

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@alia.onl/sdk",
-  "version": "4.1.0",
+  "version": "5.0.0",
   "description": "Alia AI chat SDK — reusable chat UI, voice, and streaming components for Oxy apps",
   "main": "src/index.ts",
   "react-native": "src/index.ts",
@@ -20,6 +20,10 @@
   "publishConfig": {
     "access": "public"
   },
+  "scripts": {
+    "typecheck": "tsc --noEmit",
+    "check:entries": "node scripts/check-entry-isolation.mjs"
+  },
   "dependencies": {
     "@radix-ui/react-dropdown-menu": "^2.1.20",
     "@radix-ui/react-icons": "^1.3.2",
@@ -37,7 +41,7 @@
   "peerDependencies": {
     "react": "^19.2.3",
     "react-native": "^0.86.0",
-    "@oxyhq/services": "^22.12.3",
+    "@oxyhq/services": "^23.0.1",
     "@react-native-async-storage/async-storage": "^2.2.0",
     "socket.io-client": "^4.8.3",
     "expo-font": "^57.0.0",

package/src/components/AliaChatContent.tsx

@@ -1,12 +1,14 @@
 /**
  * Shared chat content component used by both AliaChatSheet and AliaChatScreen.
  *
- * Owns all chat + voice state internally. Containers only handle
- * layout chrome (sheet modal, full-screen SafeAreaView) and pass a `header`
- * render prop that receives the live mark state.
+ * Owns the text chat; voice is an injected capability (`voiceSession`) so that
+ * nothing here reaches `livekit-client` at module-evaluation time. Containers
+ * only handle layout chrome (sheet modal, full-screen SafeAreaView) and pass a
+ * `header` render prop that receives the live mark state.
  */
 
 import React, {
+  Suspense,
   useState,
   useCallback,
   useEffect,
@@ -18,19 +20,20 @@ import { useSharedValue } from 'react-native-reanimated';
 import type { SharedValue } from 'react-native-reanimated';
 import Volume2 from 'lucide-react-native/icons/volume-2';
 import { useAliaChat, type UseAliaChatOptions } from '../hooks/useAliaChat';
-import { useVoiceRoom } from '../hooks/useVoiceRoom';
-import { useAudioLevelMonitor } from '../hooks/useAudioLevelMonitor';
-import { useAudioLevels } from '../hooks/useAudioLevels';
 import { useTTS } from '../hooks/useTTS';
 import { useAmbientWave } from '../hooks/useAmbientWave';
 import { VoiceOverlay } from './voice/VoiceOverlay';
-import { VoiceControls } from './voice/VoiceControls';
 import { AliaChatMessageList } from './AliaChatMessageList';
 import { AliaWelcomeMessage, type WelcomeSuggestion } from './AliaWelcomeMessage';
 import { PromptInput } from './ui/prompt-input/prompt-input';
 import { Button } from './ui/button';
 import type { AliaMarkState } from './AliaMark';
-import type { ChatMessage, VoiceMessage } from '../types';
+import type {
+  ChatMessage,
+  VoiceMessage,
+  VoiceSessionComponent,
+  VoiceSessionState,
+} from '../types';
 import type { Completion } from './ui/prompt-input/context';
 
 export interface AliaChatContentProps {
@@ -64,6 +67,12 @@ export interface AliaChatContentProps {
   primaryColor?: string;
   /** Dark-mode flag — forwarded to the ambient wave overlay. */
   isDarkMode?: boolean;
+  /**
+   * Voice capability — pass `VoiceSession` from `@alia.onl/sdk/voice` to offer
+   * voice calls. Omitted, the chat is text-only and no LiveKit code is reachable
+   * from this module. May be a `React.lazy` component.
+   */
+  voiceSession?: VoiceSessionComponent;
 }
 
 /** Adapt a voice message into the chat message format. */
@@ -100,6 +109,7 @@ export function AliaChatContent({
   header,
   primaryColor,
   isDarkMode,
+  voiceSession: VoiceSession,
 }: AliaChatContentProps) {
   // ── Chat ──
   const chatOptions: UseAliaChatOptions = { apiUrl, model, clientContext };
@@ -119,60 +129,36 @@ export function AliaChatContent({
   const internalScrollOffsetY = useSharedValue(0);
   const scrollOffsetY = externalScrollOffsetY ?? internalScrollOffsetY;
 
-  // ── Voice room ──
-  const voiceRoom = useVoiceRoom({ apiUrl });
-  const { captureLevel, playbackLevel } = useAudioLevelMonitor(voiceRoom.room, voiceRoom.isConnected);
-  const { waveAmplitude } = useAudioLevels({
-    captureLevel,
-    playbackLevel,
-    agentState: voiceRoom.agentState,
-    isConnected: voiceRoom.isConnected,
-  });
-
   // ── TTS ──
   const tts = useTTS({ apiUrl });
 
   // ── Voice mode state ──
+  // `isVoiceActive` mounts the session; `voiceState` is what the session reports
+  // back, and stays null until the call is under way.
   const [isVoiceActive, setIsVoiceActive] = useState(false);
+  const [voiceState, setVoiceState] = useState<VoiceSessionState | null>(null);
   const textSnapshotRef = useRef<ChatMessage[]>([]);
 
   // Merge voice messages into chat messages
+  const voiceMessages = voiceState?.messages;
   useEffect(() => {
-    if (!isVoiceActive) return;
-    const adapted = voiceRoom.messages.map(adaptVoiceMessage);
-    setMessages([...textSnapshotRef.current, ...adapted]);
-  }, [voiceRoom.messages, isVoiceActive, setMessages]);
-
-  // Auto-deactivate on error
-  useEffect(() => {
-    if (!isVoiceActive) return;
-    if (voiceRoom.error) {
-      deactivateVoice();
-    } else if (voiceRoom.roomState === 'error') {
-      deactivateVoice();
-    }
-  }, [voiceRoom.error, voiceRoom.roomState, isVoiceActive]);
-
-  // Auto-deactivate on unexpected disconnection
-  useEffect(() => {
-    if (!isVoiceActive) return;
-    if (voiceRoom.roomState === 'disconnected' && textSnapshotRef.current.length > 0) {
-      deactivateVoice();
-    }
-  }, [voiceRoom.roomState, isVoiceActive]);
+    if (!voiceMessages) return;
+    setMessages([...textSnapshotRef.current, ...voiceMessages.map(adaptVoiceMessage)]);
+  }, [voiceMessages, setMessages]);
 
   const activateVoice = useCallback(() => {
-    if (isVoiceActive || voiceRoom.roomState === 'connecting') return;
+    if (isVoiceActive) return;
     textSnapshotRef.current = [...messages];
     setIsVoiceActive(true);
-    voiceRoom.connect();
-  }, [isVoiceActive, voiceRoom, messages]);
+  }, [isVoiceActive, messages]);
 
+  // Unmounting the session disconnects the room; the transcript merged above
+  // stays in the chat.
   const deactivateVoice = useCallback(() => {
-    voiceRoom.disconnect();
     setIsVoiceActive(false);
+    setVoiceState(null);
     textSnapshotRef.current = [];
-  }, [voiceRoom]);
+  }, []);
 
   // ── TTS read aloud handler ──
   const handleReadAloud = useCallback(
@@ -183,10 +169,11 @@ export function AliaChatContent({
   );
 
   // ── Mark state (drives the header brand mark) ──
+  const voiceAgentState = voiceState?.agentState;
   const markState = useMemo<AliaMarkState>(() => {
     if (isVoiceActive) {
-      if (voiceRoom.agentState === 'speaking') return 'writing';
-      if (voiceRoom.agentState === 'thinking') return 'thinking';
+      if (voiceAgentState === 'speaking') return 'writing';
+      if (voiceAgentState === 'thinking') return 'thinking';
       return 'idle';
     }
     if (!isStreaming) return 'idle';
@@ -195,7 +182,7 @@ export function AliaChatContent({
       return 'working';
     if (lastMsg?.thinking) return 'thinking';
     return 'writing';
-  }, [messages, isStreaming, isVoiceActive, voiceRoom.agentState]);
+  }, [messages, isStreaming, isVoiceActive, voiceAgentState]);
 
   const welcomeComponent = useMemo(() => {
     if (!welcomeGreeting || messages.length > 0) return undefined;
@@ -211,29 +198,32 @@ export function AliaChatContent({
 
   // ── Ambient wave — one persistent overlay across idle/voice/TTS/STT ──
   const wave = useAmbientWave({
-    voice: {
-      isActive: isVoiceActive,
-      isConnected: voiceRoom.isConnected,
-      agentState: voiceRoom.agentState,
-      waveAmplitude,
-    },
+    voice: voiceState
+      ? {
+          isActive: isVoiceActive,
+          isConnected: voiceState.isConnected,
+          agentState: voiceState.agentState,
+          waveAmplitude: voiceState.waveAmplitude,
+        }
+      : undefined,
     isTTSPlaying: tts.playbackState === 'playing',
     ttsWaveAmplitude: tts.ttsWaveAmplitude,
     isGenerating: isStreaming,
   });
 
-  // ── Voice activate button for empty submit ──
+  // ── Voice activate button for empty submit (only if voice was provided) ──
   const voiceActivateButton = useMemo(
-    () => (
-      <Button
-        size="icon"
-        onPress={activateVoice}
-        className="h-8 w-8 rounded-full"
-      >
-        <Volume2 size={16} color="white" />
-      </Button>
-    ),
-    [activateVoice],
+    () =>
+      VoiceSession ? (
+        <Button
+          size="icon"
+          onPress={activateVoice}
+          className="h-8 w-8 rounded-full"
+        >
+          <Volume2 size={16} color="white" />
+        </Button>
+      ) : undefined,
+    [VoiceSession, activateVoice],
   );
 
   return (
@@ -269,20 +259,15 @@ export function AliaChatContent({
       />
 
       {/* Input or Voice Controls */}
-      {isVoiceActive ? (
-        <VoiceControls
-          roomState={voiceRoom.roomState}
-          agentState={voiceRoom.agentState}
-          isMuted={voiceRoom.isMuted}
-          cohostActive={voiceRoom.cohostActive}
-          currentSpeaker={voiceRoom.currentSpeaker}
-          roundComplete={voiceRoom.roundComplete}
-          onToggleMute={voiceRoom.toggleMute}
-          onEnableCohost={voiceRoom.enableCohost}
-          onDisableCohost={voiceRoom.disableCohost}
-          onContinueCohost={voiceRoom.continueCohost}
-          onEnd={deactivateVoice}
-        />
+      {VoiceSession && isVoiceActive ? (
+        /* Suspense so the session may be handed to us as a React.lazy import. */
+        <Suspense fallback={null}>
+          <VoiceSession
+            apiUrl={apiUrl}
+            onStateChange={setVoiceState}
+            onEnd={deactivateVoice}
+          />
+        </Suspense>
       ) : (
         <View style={styles.inputContainer}>
           <PromptInput

package/src/components/AliaChatScreen.tsx

@@ -11,6 +11,10 @@
  *   headerLeft={<BackButton />}
  * />
  * ```
+ *
+ * Voice is opt-in — pass `voiceSession={VoiceSession}` from
+ * `@alia.onl/sdk/voice` to offer calls. Without it the screen is text-only and
+ * `livekit-client` never enters the bundle.
  */
 
 import React from 'react';
@@ -19,6 +23,7 @@ import { useSafeAreaInsets } from 'react-native-safe-area-context';
 import { AliaChatContent } from './AliaChatContent';
 import { AliaMark } from './AliaMark';
 import type { WelcomeSuggestion } from './AliaWelcomeMessage';
+import type { VoiceSessionComponent } from '../types';
 
 export interface AliaChatScreenProps {
   /** App context injected into system prompt */
@@ -41,6 +46,8 @@ export interface AliaChatScreenProps {
   primaryColor?: string;
   /** Dark-mode flag — forwarded to the ambient wave overlay. */
   isDarkMode?: boolean;
+  /** Voice capability — `VoiceSession` from `@alia.onl/sdk/voice`. */
+  voiceSession?: VoiceSessionComponent;
 }
 
 export function AliaChatScreen({
@@ -54,6 +61,7 @@ export function AliaChatScreen({
   welcomeSuggestions,
   primaryColor,
   isDarkMode,
+  voiceSession,
 }: AliaChatScreenProps) {
   const insets = useSafeAreaInsets();
 
@@ -68,6 +76,7 @@ export function AliaChatScreen({
         welcomeSuggestions={welcomeSuggestions}
         primaryColor={primaryColor}
         isDarkMode={isDarkMode}
+        voiceSession={voiceSession}
         header={({ markState, hasMessages, clear }) => (
           <View className="flex-row items-center justify-between px-4 py-3">
             <View className="flex-row items-center gap-2.5">

package/src/components/AliaChatSheet.tsx

@@ -32,6 +32,7 @@ import { useSafeAreaInsets } from 'react-native-safe-area-context';
 import { AliaChatContent } from './AliaChatContent';
 import { AliaMark } from './AliaMark';
 import type { WelcomeSuggestion } from './AliaWelcomeMessage';
+import type { VoiceSessionComponent } from '../types';
 
 const { height: SCREEN_HEIGHT } = Dimensions.get('window');
 
@@ -58,6 +59,12 @@ export interface AliaChatSheetProps {
   primaryColor?: string;
   /** Dark-mode flag — forwarded to the ambient wave overlay (defaults to the sheet's own color scheme). */
   isDarkMode?: boolean;
+  /**
+   * Voice capability — pass `VoiceSession` from `@alia.onl/sdk/voice` to offer
+   * voice calls. Omitted, the sheet is text-only and `livekit-client` never
+   * enters the bundle.
+   */
+  voiceSession?: VoiceSessionComponent;
 }
 
 export interface AliaChatSheetRef {
@@ -66,7 +73,7 @@ export interface AliaChatSheetRef {
 }
 
 export const AliaChatSheet = forwardRef<AliaChatSheetRef, AliaChatSheetProps>(
-  ({ clientContext, model, apiUrl, welcomeGreeting, welcomeSubtitle, welcomeSuggestions, primaryColor, isDarkMode }, ref) => {
+  ({ clientContext, model, apiUrl, welcomeGreeting, welcomeSubtitle, welcomeSuggestions, primaryColor, isDarkMode, voiceSession }, ref) => {
     const scheme = useColorScheme();
     const isDark = scheme === 'dark';
     const insets = useSafeAreaInsets();
@@ -250,6 +257,7 @@ export const AliaChatSheet = forwardRef<AliaChatSheetRef, AliaChatSheetProps>(
                   welcomeSuggestions={welcomeSuggestions}
                   primaryColor={primaryColor}
                   isDarkMode={isDarkMode ?? isDark}
+                  voiceSession={voiceSession}
                   header={({ markState, hasMessages, clear }) => (
                     <View style={styles.header}>
                       <View style={styles.headerLeft}>

package/src/components/voice/VoiceSession.tsx

@@ -0,0 +1,88 @@
+/**
+ * A live LiveKit call, mounted only while voice mode is on.
+ *
+ * This is the only place the chat UI touches `livekit-client`. The chat shell
+ * receives it as the `voiceSession` prop rather than importing it, so a
+ * text-only consumer never pulls the LiveKit client into its module graph —
+ * which is the whole reason the `@alia.onl/sdk/voice` entry exists.
+ *
+ * Mounting starts the call; unmounting ends it (`useVoiceRoom` tears the room
+ * down on unmount), so the shell only has to decide whether to render it.
+ */
+
+import { useEffect, useRef } from 'react';
+import { useVoiceRoom } from '../../hooks/useVoiceRoom';
+import { useAudioLevelMonitor } from '../../hooks/useAudioLevelMonitor';
+import { useAudioLevels } from '../../hooks/useAudioLevels';
+import { VoiceControls } from './VoiceControls';
+import type { VoiceSessionProps } from '../../types';
+
+export function VoiceSession({ apiUrl, onStateChange, onEnd }: VoiceSessionProps) {
+  const {
+    room,
+    roomState,
+    agentState,
+    isMuted,
+    error,
+    messages,
+    cohostActive,
+    currentSpeaker,
+    roundComplete,
+    isConnected,
+    connect,
+    toggleMute,
+    enableCohost,
+    disableCohost,
+    continueCohost,
+  } = useVoiceRoom({ apiUrl });
+
+  const { captureLevel, playbackLevel } = useAudioLevelMonitor(room, isConnected);
+  const { waveAmplitude } = useAudioLevels({
+    captureLevel,
+    playbackLevel,
+    agentState,
+    isConnected,
+  });
+
+  // Dial out once per mount. The ref keeps a new `connect` identity from
+  // placing a second call on top of the live one.
+  const dialedRef = useRef(false);
+  useEffect(() => {
+    if (dialedRef.current) return;
+    dialedRef.current = true;
+    connect();
+  }, [connect]);
+
+  useEffect(() => {
+    onStateChange({ isConnected, agentState, waveAmplitude, messages });
+  }, [isConnected, agentState, waveAmplitude, messages, onStateChange]);
+
+  // End the call on failure, or once a room we actually reached goes away. The
+  // "reached" flag is what keeps the pre-connect `disconnected` state — which
+  // every session starts in — from ending the call before it begins.
+  const reachedRoomRef = useRef(false);
+  useEffect(() => {
+    if (roomState === 'connected') reachedRoomRef.current = true;
+    if (error || roomState === 'error') {
+      onEnd();
+      return;
+    }
+    if (roomState === 'disconnected' && reachedRoomRef.current) onEnd();
+  }, [error, roomState, onEnd]);
+
+  return (
+    <VoiceControls
+      roomState={roomState}
+      agentState={agentState}
+      isMuted={isMuted}
+      cohostActive={cohostActive}
+      currentSpeaker={currentSpeaker}
+      roundComplete={roundComplete}
+      onToggleMute={toggleMute}
+      onEnableCohost={enableCohost}
+      onDisableCohost={disableCohost}
+      onContinueCohost={continueCohost}
+      onEnd={onEnd}
+    />
+  );
+}

package/src/index.ts

@@ -40,6 +40,13 @@ export type {
   PlanStep,
 } from './types';
 
+// ── Voice capability contract (the implementation lives in `./voice`) ──
+export type {
+  VoiceSessionComponent,
+  VoiceSessionProps,
+  VoiceSessionState,
+} from './types';
+
 // ── UI Primitives (NativeWind) ──
 export { Button, buttonVariants, buttonTextVariants } from './components/ui/button';
 export type { ButtonProps } from './components/ui/button';

package/src/types.ts

@@ -1,3 +1,6 @@
+import type { ComponentType } from 'react';
+import type { SharedValue } from 'react-native-reanimated';
+
 // ── Welcome / Suggestion Types ──
 
 export interface WelcomeSuggestion {
@@ -93,3 +96,32 @@ export interface VoiceMessage {
   isStreaming: boolean;
   toolInvocations?: VoiceToolInvocation[];
 }
+
+// ── Voice Session (injected capability) ──
+
+/** Everything a live call reports back to the chat shell. */
+export interface VoiceSessionState {
+  /** The room is up and the agent is live. */
+  isConnected: boolean;
+  agentState: AgentState;
+  /** Audio amplitude channel that drives the ambient wave overlay. */
+  waveAmplitude: SharedValue<number>;
+  /** Transcript so far; the shell adapts it into the chat message list. */
+  messages: VoiceMessage[];
+}
+
+export interface VoiceSessionProps {
+  /** Alia API base URL. */
+  apiUrl?: string;
+  /** Called on every room, agent, or transcript change. */
+  onStateChange: (state: VoiceSessionState) => void;
+  /** The call is over — the user ended it, it errored, or the room dropped. */
+  onEnd: () => void;
+}
+
+/**
+ * The voice capability handed to `AliaChatScreen` / `AliaChatSheet`. Pass
+ * `VoiceSession` from `@alia.onl/sdk/voice` to enable voice calls; leave it out
+ * and the chat is text-only, with `livekit-client` absent from the module graph.
+ */
+export type VoiceSessionComponent = ComponentType<VoiceSessionProps>;

package/src/voice.ts

@@ -4,6 +4,7 @@
 // into their module graph.
 
 // ── Voice components ──
+export { VoiceSession } from './components/voice/VoiceSession';
 export { AudioWaveVisualizer } from './components/voice/AudioWaveVisualizer';
 export { VoiceOverlay } from './components/voice/VoiceOverlay';
 export { VoiceControls } from './components/voice/VoiceControls';
@@ -23,4 +24,6 @@ export type {
   AgentState,
   VoiceMessage,
   VoiceToolInvocation,
+  VoiceSessionProps,
+  VoiceSessionState,
 } from './types';