feat(sync): full multi-line paragraph & newline support with ultra-low latency 35ms mirroring

android/app/src/main/java/com/soniox/remotemic/MainActivity.kt

@@ -41,6 +41,7 @@ import okhttp3.OkHttpClient
 import okhttp3.Request
 import okhttp3.RequestBody.Companion.toRequestBody
 import org.json.JSONObject
+import java.util.concurrent.TimeUnit
 
 class MainActivity : AppCompatActivity() {
 
@@ -88,7 +89,7 @@ class MainActivity : AppCompatActivity() {
         setupUI()
         checkPermissions()
         
-        AppLogger.log("Main", "اپلیکیشن راه‌اندازی شد (طراحی ریسپانسیو و فوکوس کامل اینپوت‌باکس روی کیبورد)")
+        AppLogger.log("Main", "اپلیکیشن همگام‌سازی صوتی سانی‌اوکس راه‌اندازی شد (نسخه بهینه‌شده فوق‌سریع v5.6)")
         
         // Initialize Collaborative WebSocket Client
         streamDictationClient = StreamDictationClient(
@@ -100,12 +101,25 @@ class MainActivity : AppCompatActivity() {
                 )
             },
             onSyncStateReceived = { state ->
-                // Drop echoes and do not interrupt active user editing on phone
+                // Drop echoes originated from phone itself
                 if (state.source != "android") {
                     val timeSinceLocalEdit = System.currentTimeMillis() - lastLocalUserEditTime
                     
-                    // If user is actively typing or backspacing right now on phone, do not override with remote echo
-                    if (timeSinceLocalEdit < 1000L && binding.etTranscript.hasFocus()) {
+                    // If user is actively typing or recording right now on phone, do not override
+                    if ((timeSinceLocalEdit < 1000L && binding.etTranscript.hasFocus()) || isCurrentlyRecording) {
+                        return@StreamDictationClient
+                    }
+                    
+                    // Monotonic revision check
+                    if (state.revision > 0 && state.revision < currentRevision) {
+                        return@StreamDictationClient
+                    }
+                    
+                    // Ghost empty sync protection for opaque apps
+                    if (state.text.isEmpty() && lastLocalText.isNotEmpty() && state.source == "mac") {
+                        if (state.app.isNotEmpty() && state.app != "Desktop" && state.app != "App") {
+                            binding.tvMacStatus.text = "متصل به ${state.app} 🖥️"
+                        }
                         return@StreamDictationClient
                     }
                     
@@ -137,15 +151,17 @@ class MainActivity : AppCompatActivity() {
                 binding.tvInstruction.setTextColor(ContextCompat.getColor(this, R.color.accent_blue))
             },
             onAudioLevel = { level ->
-                val scale = 1.0f + (level * 0.35f)
-                binding.viewGlow.scaleX = scale
-                binding.viewGlow.scaleY = scale
+                if (isCurrentlyRecording) {
+                    val scale = 1.0f + (level * 0.22f)
+                    binding.viewGlow.scaleX = scale
+                    binding.viewGlow.scaleY = scale
+                }
             },
             onSpeechCompleted = { finalText ->
                 vibrate(100)
                 if (finalText.isNotEmpty()) {
                     insertSpeechAtCursor(finalText)
-                    binding.tvInstruction.text = "✨ گفتار در نشانگر درج و با مک همگام شد"
+                    binding.tvInstruction.text = "✨ گفتار با مک همگام شد"
                     binding.tvInstruction.setTextColor(ContextCompat.getColor(this, R.color.accent_green))
                 } else {
                     binding.tvInstruction.text = "صدایی تشخیص داده نشد"
@@ -160,18 +176,15 @@ class MainActivity : AppCompatActivity() {
     }
 
     /**
-     * Dual-engine keyboard visibility detector (WindowInsets + OnGlobalLayoutListener)
-     * Guarantees 100% detection on all Android versions and keyboards.
+     * Dual-engine keyboard visibility detector
      */
     private fun setupKeyboardVisibilityDetection() {
-        // Engine 1: Modern WindowInsets
         ViewCompat.setOnApplyWindowInsetsListener(binding.rootLayout) { _, insets ->
             val imeVisible = insets.isVisible(WindowInsetsCompat.Type.ime())
             updateKeyboardUIMode(imeVisible)
             insets
         }
 
-        // Engine 2: Global Layout Frame Calculation (Fallback for OEM soft keyboards)
         binding.rootLayout.viewTreeObserver.addOnGlobalLayoutListener(object : ViewTreeObserver.OnGlobalLayoutListener {
             private val r = Rect()
             override fun onGlobalLayout() {
@@ -189,14 +202,11 @@ class MainActivity : AppCompatActivity() {
         isKeyboardCurrentlyVisible = isKeyboardOpen
 
         if (isKeyboardOpen) {
-            // KEYBOARD OPEN: Hide big mic circle, hide action buttons & divider completely!
-            // Give 100% of available space above keyboard exclusively to the huge text editor box!
             binding.bottomVoiceSection.visibility = View.GONE
             binding.actionDivider.visibility = View.GONE
             binding.actionButtonsRow.visibility = View.GONE
             binding.tvSubtitle.visibility = View.GONE
         } else {
-            // KEYBOARD CLOSED: Restore spacious layout with large glowing mic button & action bar
             binding.bottomVoiceSection.visibility = View.VISIBLE
             binding.actionDivider.visibility = View.VISIBLE
             binding.actionButtonsRow.visibility = View.VISIBLE
@@ -205,6 +215,13 @@ class MainActivity : AppCompatActivity() {
     }
 
     private fun insertSpeechAtCursor(speechText: String) {
+        val trimmedSpeech = speechText.trim()
+        
+        // Suppress empty strings or lone quotes/punctuation marks
+        if (trimmedSpeech.isEmpty() || trimmedSpeech.matches("^[\\s«»\\.\\,\\،\\؛\\؟\\!\\?\\:\\;\\-\\–—\\\"\\'\\(\\)\\[\\]\\{\\}]+$".toRegex())) {
+            return
+        }
+
         val current = binding.etTranscript.text?.toString() ?: ""
         val start = voiceInsertionCursorStart.coerceIn(0, current.length)
         val end = voiceInsertionCursorEnd.coerceIn(0, current.length)
@@ -212,15 +229,24 @@ class MainActivity : AppCompatActivity() {
         val prefix = if (start > 0) current.substring(0, start) else ""
         val suffix = if (end < current.length) current.substring(end) else ""
         
-        val formattedSpeech = if (prefix.isNotEmpty() && !prefix.endsWith(" ") && !speechText.startsWith(" ")) {
-            " $speechText"
-        } else {
-            speechText
+        val needsPreSpace = prefix.isNotEmpty() && !prefix.endsWith(" ") && !prefix.endsWith("\n")
+        val needsPostSpace = suffix.isNotEmpty() && !suffix.startsWith(" ") && !suffix.startsWith("\n") &&
+                !suffix.startsWith(",") && !suffix.startsWith("،") && !suffix.startsWith(".") &&
+                !suffix.startsWith("؟") && !suffix.startsWith("!") && !suffix.startsWith(":")
+        
+        val formattedSpeech = buildString {
+            if (needsPreSpace) append(" ")
+            append(trimmedSpeech)
+            if (needsPostSpace) append(" ")
         }
         
         val mergedText = "$prefix$formattedSpeech$suffix"
         val newCursor = (start + formattedSpeech.length).coerceIn(0, mergedText.length)
         
+        // Cancel any pending debounced sync
+        pendingSyncRunnable?.let { debounceHandler.removeCallbacks(it) }
+        pendingSyncRunnable = null
+
         isApplyingRemoteUpdate = true
         lastLocalText = mergedText
         lastLocalUserEditTime = System.currentTimeMillis()
@@ -228,9 +254,10 @@ class MainActivity : AppCompatActivity() {
         binding.etTranscript.setSelection(newCursor)
         isApplyingRemoteUpdate = false
         
-        AppLogger.log("Main", "تزریق گفتار در نشانگر: '$speechText' (موقعیت جدید: $newCursor)")
+        AppLogger.log("Main", "تزریق گفتار در نشانگر: '$trimmedSpeech' (موقعیت جدید: $newCursor)")
         
-        streamDictationClient?.sendLocalSyncState(mergedText, newCursor)
+        // Broadcast full mirrored text to Mac
+        streamDictationClient?.sendPhoneEdit(mergedText, newCursor)
     }
 
     private fun setupUI() {
@@ -247,25 +274,36 @@ class MainActivity : AppCompatActivity() {
                     lastLocalUserEditTime = System.currentTimeMillis()
                     val cur = binding.etTranscript.selectionStart.coerceIn(0, text.length)
                     
-                    // Debounce sending to Mac by 60ms to allow 120Hz lag-free backspacing and typing
+                    // Immediate sync for newlines (\n), crisp 35ms debounce for general typing
+                    val isNewlineEdit = count == 1 && s?.subSequence(start, start + count)?.contains('\n') == true
+                    val delayMs = if (isNewlineEdit) 0L else 35L
+                    
                     pendingSyncRunnable?.let { debounceHandler.removeCallbacks(it) }
                     pendingSyncRunnable = Runnable {
-                        streamDictationClient?.sendLocalSyncState(text, cur)
+                        streamDictationClient?.sendPhoneEdit(text, cur)
+                    }
+                    if (delayMs == 0L) {
+                        debounceHandler.post(pendingSyncRunnable!!)
+                    } else {
+                        debounceHandler.postDelayed(pendingSyncRunnable!!, delayMs)
                     }
-                    debounceHandler.postDelayed(pendingSyncRunnable!!, 60)
                 }
             }
             override fun afterTextChanged(s: Editable?) {}
         })
 
-        // Clear Button
+        // Clear Button (🗑️ پاک‌کردن)
         binding.btnClearText.setOnClickListener {
             isApplyingRemoteUpdate = true
             binding.etTranscript.setText("")
             lastLocalText = ""
             lastLocalUserEditTime = System.currentTimeMillis()
             isApplyingRemoteUpdate = false
-            streamDictationClient?.sendLocalSyncState("", 0)
+            
+            pendingSyncRunnable?.let { debounceHandler.removeCallbacks(it) }
+            pendingSyncRunnable = null
+            
+            streamDictationClient?.sendPhoneEdit("", 0)
             binding.tvInstruction.text = getString(R.string.hold_to_speak)
             binding.tvInstruction.setTextColor(ContextCompat.getColor(this, R.color.text_secondary))
         }
@@ -293,8 +331,8 @@ class MainActivity : AppCompatActivity() {
             binding.tvInstruction.text = "در حال درج متن در مک..."
             binding.tvInstruction.setTextColor(ContextCompat.getColor(this, R.color.accent_yellow))
 
-            // 1. Direct WebSocket broadcast
-            streamDictationClient?.sendLocalSyncState(text, cur)
+            // 1. Direct WebSocket broadcast with force_replace
+            streamDictationClient?.sendForceReplace(text, cur)
 
             // 2. Direct HTTP Post guarantee
             lifecycleScope.launch {
@@ -323,12 +361,16 @@ class MainActivity : AppCompatActivity() {
         return when (event.action) {
             MotionEvent.ACTION_DOWN -> {
                 if (checkAudioPermission()) {
+                    // Fast sync: ensure socket is connected immediately
+                    streamDictationClient?.connectWebSocket()
+                    
                     val selStart = binding.etTranscript.selectionStart
                     val selEnd = binding.etTranscript.selectionEnd
-                    val totalLen = binding.etTranscript.text?.length ?: 0
+                    val currentText = binding.etTranscript.text?.toString() ?: ""
+                    val totalLen = currentText.length
                     
-                    voiceInsertionCursorStart = if (selStart >= 0) selStart else totalLen
-                    voiceInsertionCursorEnd = if (selEnd >= 0) selEnd else totalLen
+                    voiceInsertionCursorStart = if (selStart in 0..totalLen) selStart else totalLen
+                    voiceInsertionCursorEnd = if (selEnd in 0..totalLen) selEnd else totalLen
                     
                     startRecording()
                 }
@@ -336,6 +378,16 @@ class MainActivity : AppCompatActivity() {
             }
             MotionEvent.ACTION_UP, MotionEvent.ACTION_CANCEL -> {
                 if (isCurrentlyRecording) {
+                    // Update insertion point to where cursor was right before processing
+                    val selStart = binding.etTranscript.selectionStart
+                    val selEnd = binding.etTranscript.selectionEnd
+                    val currentText = binding.etTranscript.text?.toString() ?: ""
+                    val totalLen = currentText.length
+                    
+                    if (selStart in 0..totalLen && selEnd in 0..totalLen) {
+                        voiceInsertionCursorStart = selStart
+                        voiceInsertionCursorEnd = selEnd
+                    }
                     stopRecordingAndProcess()
                 }
                 true
@@ -346,7 +398,11 @@ class MainActivity : AppCompatActivity() {
 
     private suspend fun sendDirectPaste(text: String, cursor: Int): Boolean {
         return try {
-            val client = OkHttpClient()
+            val client = OkHttpClient.Builder()
+                .connectTimeout(2500, TimeUnit.MILLISECONDS)
+                .writeTimeout(3000, TimeUnit.MILLISECONDS)
+                .readTimeout(3000, TimeUnit.MILLISECONDS)
+                .build()
             val json = JSONObject().apply {
                 put("text", text)
                 put("cursor_pos", cursor)
@@ -354,13 +410,25 @@ class MainActivity : AppCompatActivity() {
             }.toString()
             val mediaType = "application/json; charset=utf-8".toMediaType()
             val body = json.toRequestBody(mediaType)
-            val req = Request.Builder()
-                .url("http://$gatewayHost/paste")
-                .post(body)
-                .build()
-            client.newCall(req).execute().use { resp ->
-                resp.isSuccessful
+            
+            val hosts = listOf(gatewayHost, "116.16.16.19:8999", "2.180.16.250:8999").distinct()
+            var success = false
+            for (h in hosts) {
+                try {
+                    val req = Request.Builder()
+                        .url("http://$h/paste")
+                        .post(body)
+                        .build()
+                    client.newCall(req).execute().use { resp ->
+                        if (resp.isSuccessful) {
+                            success = true
+                            return@use
+                        }
+                    }
+                    if (success) break
+                } catch (_: Exception) {}
             }
+            success
         } catch (e: Exception) {
             AppLogger.log("Main", "خطای ارسال دستی: ${e.message}")
             false
@@ -456,12 +524,12 @@ class MainActivity : AppCompatActivity() {
     }
 
     private fun startPulseAnimation() {
-        val scaleX = PropertyValuesHolder.ofFloat(View.SCALE_X, 1.0f, 1.25f, 1.0f)
-        val scaleY = PropertyValuesHolder.ofFloat(View.SCALE_Y, 1.0f, 1.25f, 1.0f)
-        val alpha = PropertyValuesHolder.ofFloat(View.ALPHA, 0.3f, 0.7f, 0.3f)
+        val scaleX = PropertyValuesHolder.ofFloat(View.SCALE_X, 1.0f, 1.18f, 1.0f)
+        val scaleY = PropertyValuesHolder.ofFloat(View.SCALE_Y, 1.0f, 1.18f, 1.0f)
+        val alpha = PropertyValuesHolder.ofFloat(View.ALPHA, 0.4f, 0.85f, 0.4f)
         
         pulseAnimator = ObjectAnimator.ofPropertyValuesHolder(binding.viewGlow, scaleX, scaleY, alpha).apply {
-            duration = 1000
+            duration = 900
             repeatCount = ValueAnimator.INFINITE
             interpolator = AccelerateDecelerateInterpolator()
             start()
@@ -472,7 +540,7 @@ class MainActivity : AppCompatActivity() {
         pulseAnimator?.cancel()
         binding.viewGlow.scaleX = 1.0f
         binding.viewGlow.scaleY = 1.0f
-        binding.viewGlow.alpha = 0.3f
+        binding.viewGlow.alpha = 0.0f
     }
 
     private fun vibrate(durationMs: Long) {

android/app/src/main/java/com/soniox/remotemic/StreamDictationClient.kt

@@ -44,18 +44,26 @@ class StreamDictationClient(
     private val isRecording = AtomicBoolean(false)
     private val mainHandler = Handler(Looper.getMainLooper())
 
+    private val candidateHosts: List<String> = listOf(
+        host,
+        if (host.contains("116.16.16.19")) "2.180.16.250:8999" else "116.16.16.19:8999"
+    ).filter { it.isNotBlank() }.distinct()
+    private var currentHostIndex = 0
+
     private val okHttpClient = OkHttpClient.Builder()
-        .connectTimeout(5, TimeUnit.SECONDS)
+        .connectTimeout(3500, TimeUnit.MILLISECONDS)
         .readTimeout(0, TimeUnit.MILLISECONDS) // Persistent WebSocket
-        .writeTimeout(5, TimeUnit.SECONDS)
-        .pingInterval(10, TimeUnit.SECONDS)
+        .writeTimeout(5000, TimeUnit.MILLISECONDS)
+        .pingInterval(4, TimeUnit.SECONDS) // 4s active heartbeat to keep NAT tables alive
         .retryOnConnectionFailure(true)
         .build()
 
-    private var webSocket: WebSocket? = null
+    private var activeWebSocket: WebSocket? = null
     private val isConnected = AtomicBoolean(false)
+    private val isConnecting = AtomicBoolean(false)
     private var currentSessionId: String = ""
     private var isSessionActive = AtomicBoolean(false)
+    private var retryAttempt = 0
 
     init {
         connectWebSocket()
@@ -63,21 +71,30 @@ class StreamDictationClient(
 
     @Synchronized
     fun connectWebSocket() {
-        if (isConnected.get() && webSocket != null) return
+        if (isConnected.get() && activeWebSocket != null) return
+        if (isConnecting.getAndSet(true)) return
 
-        val cleanHost = host.removePrefix("http://").removePrefix("https://").removePrefix("ws://").removePrefix("wss://")
+        val targetHost = candidateHosts[currentHostIndex % candidateHosts.size]
+        val cleanHost = targetHost.removePrefix("http://").removePrefix("https://").removePrefix("ws://").removePrefix("wss://")
         val wsUrl = "ws://$cleanHost/ws/stream"
-        AppLogger.log(tag, "اتصال به سوکت همگام‌سازی بلادرنگ: $wsUrl")
+        AppLogger.log(tag, "اتصال به سوکت همگام‌سازی بلادرنگ ($targetHost): $wsUrl")
+
+        val req = Request.Builder()
+            .url(wsUrl)
+            .header("User-Agent", "SonioxAndroidRemote/5.6")
+            .build()
 
-        val req = Request.Builder().url(wsUrl).build()
-        webSocket = okHttpClient.newWebSocket(req, object : WebSocketListener() {
-            override fun onOpen(ws: WebSocket, response: Response) {
-                AppLogger.log(tag, "🟢 سوکت همگام‌سازی بلادرنگ متصل شد")
+        activeWebSocket = okHttpClient.newWebSocket(req, object : WebSocketListener() {
+            override fun onOpen(webSocket: WebSocket, response: Response) {
+                val connectedHost = candidateHosts[currentHostIndex % candidateHosts.size]
+                AppLogger.log(tag, "🟢 سوکت همگام‌سازی بلادرنگ متصل شد ($connectedHost)")
                 isConnected.set(true)
+                isConnecting.set(false)
+                retryAttempt = 0
                 mainHandler.post { onConnectionStateChanged(true) }
             }
 
-            override fun onMessage(ws: WebSocket, text: String) {
+            override fun onMessage(webSocket: WebSocket, text: String) {
                 try {
                     val json = JSONObject(text)
                     val type = json.optString("type")
@@ -121,40 +138,67 @@ class StreamDictationClient(
                 }
             }
 
-            override fun onFailure(ws: WebSocket, t: Throwable, response: Response?) {
+            override fun onFailure(webSocket: WebSocket, t: Throwable, response: Response?) {
                 AppLogger.log(tag, "🔴 قطع سوکت: ${t.message}. تلاش مجدد...")
-                isConnected.set(false)
-                webSocket = null
-                mainHandler.post { onConnectionStateChanged(false) }
-                mainHandler.postDelayed({ connectWebSocket() }, 2000)
+                handleDisconnect()
             }
 
-            override fun onClosed(ws: WebSocket, code: Int, reason: String) {
+            override fun onClosed(webSocket: WebSocket, code: Int, reason: String) {
                 AppLogger.log(tag, "سوکت بسته شد ($code). تلاش مجدد...")
+                handleDisconnect()
+            }
+
+            private fun handleDisconnect() {
                 isConnected.set(false)
-                webSocket = null
+                isConnecting.set(false)
+                activeWebSocket = null
                 mainHandler.post { onConnectionStateChanged(false) }
-                mainHandler.postDelayed({ connectWebSocket() }, 2000)
+                
+                // Fast failover between LAN and WAN
+                currentHostIndex++
+                retryAttempt++
+                val delayMs = if (retryAttempt <= 2) 350L else minOf(500L * (1L shl minOf(retryAttempt, 2)), 2000L)
+                mainHandler.postDelayed({ connectWebSocket() }, delayMs)
             }
         })
     }
 
     /**
-     * Broadcasts phone's updated text & cursor to Mac in sub-15ms
+     * Broadcasts phone's live text to Mac for immediate real-time mirroring
+     */
+    fun sendPhoneEdit(text: String, cursor: Int, selection: Int = 0) {
+        if (!isConnected.get() || activeWebSocket == null) {
+            connectWebSocket()
+        }
+        val payload = JSONObject().apply {
+            put("type", "phone_input_edit")
+            put("action", "phone_input_edit")
+            put("source", "android")
+            put("text", text)
+            put("cursor", cursor)
+            put("selection", selection)
+            put("timestamp", System.currentTimeMillis() / 1000.0)
+        }.toString()
+        activeWebSocket?.send(payload)
+    }
+
+    /**
+     * Broadcasts phone's full updated text & cursor to Mac for replace (Cmd+A + Cmd+V)
      */
-    fun sendLocalSyncState(text: String, cursor: Int, selection: Int = 0) {
-        if (!isConnected.get() || webSocket == null) {
+    fun sendForceReplace(text: String, cursor: Int, selection: Int = 0) {
+        if (!isConnected.get() || activeWebSocket == null) {
             connectWebSocket()
         }
         val payload = JSONObject().apply {
-            put("type", "sync_state")
+            put("type", "update_input")
+            put("action", "update_input")
             put("source", "android")
             put("text", text)
             put("cursor", cursor)
             put("selection", selection)
             put("timestamp", System.currentTimeMillis() / 1000.0)
         }.toString()
-        webSocket?.send(payload)
+        activeWebSocket?.send(payload)
     }
 
     @SuppressLint("MissingPermission")
@@ -165,7 +209,7 @@ class StreamDictationClient(
         isRecording.set(true)
         isSessionActive.set(true)
 
-        if (!isConnected.get() || webSocket == null) {
+        if (!isConnected.get() || activeWebSocket == null) {
             connectWebSocket()
         }
 
@@ -174,7 +218,7 @@ class StreamDictationClient(
             put("session_id", currentSessionId)
             put("cursor_pos", cursorPos)
         }.toString()
-        webSocket?.send(startFrame)
+        activeWebSocket?.send(startFrame)
         AppLogger.log(tag, "🎙️ شروع استریم گفتار در موقعیت نشانگر $cursorPos...")
 
         val minBufferSize = AudioRecord.getMinBufferSize(sampleRate, channelConfig, audioFormat)
@@ -207,7 +251,7 @@ class StreamDictationClient(
                     val bytesRead = audioRecord?.read(chunk, 0, chunk.size) ?: -1
                     if (bytesRead > 0) {
                         val slice = if (bytesRead == chunk.size) chunk.clone() else chunk.copyOf(bytesRead)
-                        webSocket?.send(slice.toByteString())
+                        activeWebSocket?.send(slice.toByteString())
 
                         var sum = 0.0
                         val samplesCount = bytesRead / 2
@@ -255,7 +299,7 @@ class StreamDictationClient(
             put("session_id", currentSessionId)
             put("cursor_pos", cursorPos)
         }.toString()
-        webSocket?.send(stopFrame)
+        activeWebSocket?.send(stopFrame)
         AppLogger.log(tag, "⏹️ پایان ضبط. دریافت متن نهایی...")
     }
 
@@ -263,7 +307,7 @@ class StreamDictationClient(
         isRecording.set(false)
         try {
             audioRecord?.release()
-            webSocket?.close(1000, "Client Shutdown")
+            activeWebSocket?.close(1000, "Client Shutdown")
         } catch (e: Exception) {}
     }
 }

android/app/src/main/res/drawable/bg_glow.xml

@@ -0,0 +1,14 @@
+<?xml version="1.0" encoding="utf-8"?>
+<shape xmlns:android="http://schemas.android.com/apk/res/android"
+    android:shape="oval">
+    <gradient
+        android:angle="0"
+        android:endColor="#003B82F6"
+        android:centerColor="#263B82F6"
+        android:startColor="#663B82F6"
+        android:type="radial"
+        android:gradientRadius="70dp" />
+    <size
+        android:width="140dp"
+        android:height="140dp" />
+</shape>

android/app/src/main/res/drawable/bg_mic_button.xml

@@ -1,13 +1,17 @@
 <?xml version="1.0" encoding="utf-8"?>
 <ripple xmlns:android="http://schemas.android.com/apk/res/android"
-    android:color="#40FFFFFF">
+    android:color="#33000000">
     <item>
         <shape android:shape="oval">
+            <!-- Modern Apple-Style Frost / Titanium Gradient Backdrop -->
             <gradient
                 android:angle="45"
-                android:endColor="#2563EB"
-                android:startColor="#3B82F6"
+                android:endColor="#CBD5E1"
+                android:startColor="#F1F5F9"
                 android:type="linear" />
+            <stroke
+                android:width="2.5dp"
+                android:color="#94A3B8" />
             <size
                 android:width="140dp"
                 android:height="140dp" />

android/app/src/main/res/drawable/bg_mic_button_active.xml

@@ -6,6 +6,9 @@
         android:endColor="#DC2626"
         android:startColor="#F43F5E"
         android:type="linear" />
+    <stroke
+        android:width="3dp"
+        android:color="#FECDD3" />
     <size
         android:width="140dp"
         android:height="140dp" />

android/app/src/main/res/drawable/ic_mic_mac.xml

@@ -0,0 +1,42 @@
+<vector xmlns:android="http://schemas.android.com/apk/res/android"
+    android:width="64dp"
+    android:height="64dp"
+    android:viewportWidth="64"
+    android:viewportHeight="64">
+
+    <!-- Microphone Capsule Base & Stand (Deep Sleek Black / Slate) -->
+    <!-- Mic Top Head / Capsule -->
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M32,6 C26.48,6 22,10.48 22,16 L22,30 C22,35.52 26.48,40 32,40 C37.52,40 42,35.52 42,30 L42,16 C42,10.48 37.52,6 32,6 Z" />
+
+    <!-- Mic Mesh Highlight / Inner Grille Lines -->
+    <path
+        android:fillColor="#334155"
+        android:pathData="M24,20 L40,20 L40,22 L24,22 Z M24,15 L40,15 L40,17 L24,17 Z M26,10 L38,10 L38,12 L26,12 Z" />
+
+    <!-- Outer Mic Cradle / U-Bracket -->
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M48,27 C48,36.2 40.7,43.7 34.5,44.8 L34.5,52 L40,52 C41.1,52 42,52.9 42,54 C42,55.1 41.1,56 40,56 L24,56 C22.9,56 22,55.1 22,54 C22,52.9 22.9,52 24,52 L29.5,52 L29.5,44.8 C23.3,43.7 16,36.2 16,27 L20,27 C20,34 25.4,39.5 32,39.5 C38.6,39.5 44,34 44,27 L48,27 Z" />
+
+    <!-- Mac Desktop / Display Badge Accent on Mic Body -->
+    <!-- Mini Mac Display / Apple Screen Symbol in Sleek Silver Accent -->
+    <path
+        android:fillColor="#38BDF8"
+        android:pathData="M28,26 L36,26 C37.1,26 38,26.9 38,28 L38,33 C38,34.1 37.1,35 36,35 L28,35 C26.9,35 26,34.1 26,33 L26,28 C26,26.9 26.9,26 28,26 Z" />
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M29,28 L35,28 L35,32 L29,32 Z" />
+    <path
+        android:fillColor="#38BDF8"
+        android:pathData="M30,35 L34,35 L35,37 L29,37 Z" />
+
+    <!-- Apple Leaf Accent at Top Corner -->
+    <path
+        android:fillColor="#10B981"
+        android:pathData="M48,10 C48,7.2 46.2,5.2 44,5 C43.8,7.4 45.4,9.8 48,10 Z" />
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M49.5,10.2 C48.3,10.2 47.3,10.9 46.5,10.9 C45.7,10.9 44.8,10.3 43.6,10.3 C41.6,10.3 39.8,12 39.8,15.2 C39.8,19.2 43.2,24 46.5,24 C47.5,24 48.2,23.3 49.3,23.3 C50.4,23.3 51,24 52.1,24 C54.3,24 56.5,20.8 57.1,19.5 C53.4,17.7 53,12.5 56.6,10.8 C55.3,8.9 53.3,8.7 52.4,8.7 C51.2,8.7 50.3,9.5 49.5,10.2 Z" />
+</vector>

android/app/src/main/res/drawable/ic_mic_minimal.xml

@@ -0,0 +1,16 @@
+<vector xmlns:android="http://schemas.android.com/apk/res/android"
+    android:width="48dp"
+    android:height="48dp"
+    android:viewportWidth="24"
+    android:viewportHeight="24">
+
+    <!-- Minimalist Deep Matte Black Microphone Body -->
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M12,14 C13.66,14 15,12.66 15,11 L15,5 C15,3.34 13.66,2 12,2 C10.34,2 9,3.34 9,5 L9,11 C9,12.66 10.34,14 12,14 Z" />
+
+    <!-- Minimalist Outer Base Stand & Cradle -->
+    <path
+        android:fillColor="#0F172A"
+        android:pathData="M17.3,11 C17.3,13.93 14.93,16.3 12,16.3 C9.07,16.3 6.7,13.93 6.7,11 L5,11 C5,14.49 7.72,17.36 11,17.85 L11,21 L13,21 L13,17.85 C16.28,17.36 19,14.49 19,11 L17.3,11 Z" />
+</vector>

android/app/src/main/res/layout/activity_main.xml

@@ -6,6 +6,8 @@
     android:layout_width="match_parent"
     android:layout_height="match_parent"
     android:background="@color/bg_dark"
+    android:clipChildren="false"
+    android:clipToPadding="false"
     android:padding="12dp">
 
     <!-- Header Section -->
@@ -79,7 +81,7 @@
             android:background="@drawable/bg_card"
             android:contentDescription="لاگ‌ها"
             android:padding="8dp"
-            android:src="@drawable/ic_mic"
+            android:src="@drawable/ic_mic_minimal"
             app:tint="@color/accent_blue" />
     </LinearLayout>
 
@@ -89,7 +91,7 @@
         android:layout_width="match_parent"
         android:layout_height="0dp"
         android:layout_marginTop="8dp"
-        android:layout_marginBottom="8dp"
+        android:layout_marginBottom="16dp"
         android:background="@drawable/bg_card"
         android:orientation="vertical"
         android:padding="12dp"
@@ -133,7 +135,7 @@
                 android:textSize="11sp" />
         </LinearLayout>
 
-        <!-- Fully Editable Multi-line EditText (Expands to fill entire space) -->
+        <!-- Fully Editable Multi-line EditText -->
         <EditText
             android:id="@+id/etTranscript"
             android:layout_width="match_parent"
@@ -194,11 +196,14 @@
         </LinearLayout>
     </LinearLayout>
 
-    <!-- Bottom Voice Section (Hidden when soft keyboard opens) -->
+    <!-- Bottom Voice Section (Centered with generous breathing space) -->
     <LinearLayout
         android:id="@+id/bottomVoiceSection"
         android:layout_width="match_parent"
         android:layout_height="wrap_content"
+        android:layout_marginBottom="8dp"
+        android:clipChildren="false"
+        android:clipToPadding="false"
         android:gravity="center_horizontal"
         android:orientation="vertical"
         app:layout_constraintBottom_toBottomOf="parent">
@@ -206,35 +211,39 @@
         <!-- Center Glowing Mic Button Container -->
         <FrameLayout
             android:id="@+id/micContainer"
-            android:layout_width="120dp"
-            android:layout_height="120dp"
-            android:layout_marginBottom="6dp">
+            android:layout_width="140dp"
+            android:layout_height="140dp"
+            android:layout_marginBottom="8dp"
+            android:clipChildren="false"
+            android:clipToPadding="false">
 
-            <!-- Outer Pulsing Glow -->
+            <!-- Outer Pulsing Glow Ambient Ring -->
             <View
                 android:id="@+id/viewGlow"
-                android:layout_width="120dp"
-                android:layout_height="120dp"
+                android:layout_width="140dp"
+                android:layout_height="140dp"
                 android:layout_gravity="center"
-                android:alpha="0.3"
-                android:background="@drawable/bg_mic_button" />
+                android:alpha="0.0"
+                android:background="@drawable/bg_glow" />
 
-            <!-- Main Touch Button -->
+            <!-- Main Touch Button (Matte Frost Titanium Circle) -->
             <FrameLayout
                 android:id="@+id/btnMic"
-                android:layout_width="95dp"
-                android:layout_height="95dp"
+                android:layout_width="96dp"
+                android:layout_height="96dp"
                 android:layout_gravity="center"
                 android:background="@drawable/bg_mic_button"
                 android:clickable="true"
+                android:elevation="4dp"
                 android:focusable="true">
 
+                <!-- Minimalist Sleek Matte Black Mic Icon -->
                 <ImageView
                     android:id="@+id/ivMicIcon"
-                    android:layout_width="40dp"
-                    android:layout_height="40dp"
+                    android:layout_width="48dp"
+                    android:layout_height="48dp"
                     android:layout_gravity="center"
-                    android:src="@drawable/ic_mic" />
+                    android:src="@drawable/ic_mic_minimal" />
             </FrameLayout>
         </FrameLayout>
 
@@ -243,7 +252,7 @@
             android:id="@+id/tvInstruction"
             android:layout_width="wrap_content"
             android:layout_height="wrap_content"
-            android:layout_marginBottom="4dp"
+            android:layout_marginBottom="6dp"
             android:text="@string/hold_to_speak"
             android:textColor="@color/text_secondary"
             android:textSize="13sp"

mac/package.sh

@@ -25,8 +25,8 @@ cp "$BUILD_DIR/SonioxVoice" "$APP_BUNDLE/Contents/MacOS/SonioxVoice"
 cp "$PROJECT_DIR/Info.plist" "$APP_BUNDLE/Contents/Info.plist"
 cp "$PROJECT_DIR/resources/AppIcon.icns" "$APP_BUNDLE/Contents/Resources/AppIcon.icns"
 
-# 3. Ad-hoc Codesign
-codesign --force --deep --sign - "$APP_BUNDLE"
+# 3. Stable Codesign with explicit identifier
+codesign --force --deep --sign - --identifier "com.soniox.voice" "$APP_BUNDLE"
 
 echo "✅ App bundle assembled at $APP_BUNDLE"
 
@@ -52,4 +52,3 @@ rm -rf "/Applications/$APP_NAME.app"
 cp -R "$APP_BUNDLE" "/Applications/$APP_NAME.app"
 
 echo "🎉 All Done Successfully!"
-ls -lh "$BUILD_DIR"

mac/src/AppDelegate.swift

@@ -10,6 +10,7 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
     private var isBusyFinalizing = false
     private var currentAudioLevel: Float = 0.0
     private var latestPartialText: String? = nil
+    private var permissionPollTimer: Timer?
     
     // Remote Android Phone Local Receiver (Port 8999)
     private var remoteListener: NWListener?
@@ -60,14 +61,20 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
         startRemotePasteServer()
         
         // Connect to Linux Persistent Gateway (via local tunnel 18999 -> 8999)
-        RelayClient.shared.onRemoteUpdateReceived = { [weak self] text, cursor, isFullReplace in
-            guard let self = self else { return }
-            print("AppDelegate: 📥 Clean Remote Input Update (Silent): '\(text.prefix(30))...' (replace: \(isFullReplace))")
-            HUDOverlayController.shared.show(state: .success(text: text))
+        RelayClient.shared.onRemoteUpdateReceived = { text, cursor, isFullReplace in
+            print("AppDelegate: 📥 Clean Remote Input Update: '\(text.prefix(30))...' (replace: \(isFullReplace))")
+            
+            // 1. FIRST apply the remote update directly into the active input while window is in pristine focus!
             FocusedInputSync.shared.applyRemoteUpdate(text: text, cursor: cursor, isFullReplace: isFullReplace)
+            
+            // 2. THEN show the floating success HUD notification
+            DispatchQueue.main.asyncAfter(deadline: .now() + 0.05) {
+                HUDOverlayController.shared.show(state: .success(text: text))
+            }
         }
         RelayClient.shared.start()
         
+        HotkeyManager.shared.registerHotkeys()
         checkInitialPermissions()
     }
     
@@ -91,8 +98,8 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
     }
     
     private func handleRemoteConnection(_ connection: NWConnection) {
-        connection.receive(minimumIncompleteLength: 1, maximumLength: 65536) { [weak self] data, _, _, _ in
-            guard let self = self, let data = data, let reqStr = String(data: data, encoding: .utf8) else {
+        connection.receive(minimumIncompleteLength: 1, maximumLength: 65536) { data, _, _, _ in
+            guard let data = data, let reqStr = String(data: data, encoding: .utf8) else {
                 connection.cancel()
                 return
             }
@@ -130,6 +137,37 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
     }
     
     private func checkInitialPermissions() {
+        // 1. Force macOS to trigger the system Accessibility prompt dialog!
+        let axOptions = [kAXTrustedCheckOptionPrompt.takeUnretainedValue() as String: true] as CFDictionary
+        let axTrusted = AXIsProcessTrustedWithOptions(axOptions)
+        print("AppDelegate: 🔒 Accessibility Permission Status (trusted: \(axTrusted))")
+        
+        if !axTrusted {
+            print("AppDelegate: ⚠️ Accessibility not yet granted! Prompting user and monitoring...")
+            // Open Accessibility pane directly on both modern and legacy macOS
+            DispatchQueue.main.asyncAfter(deadline: .now() + 0.5) {
+                if let url1 = URL(string: "x-apple.systempreferences:com.apple.settings.PrivacySecurity.extension?Privacy_Accessibility") {
+                    NSWorkspace.shared.open(url1)
+                }
+                if let url2 = URL(string: "x-apple.systempreferences:com.apple.preference.security?Privacy_Accessibility") {
+                    NSWorkspace.shared.open(url2)
+                }
+            }
+            
+            // Poll every 1.0 seconds until user toggles the switch
+            permissionPollTimer = Timer.scheduledTimer(withTimeInterval: 1.0, repeats: true) { [weak self] timer in
+                let nowTrusted = AXIsProcessTrusted()
+                if nowTrusted {
+                    print("AppDelegate: 🟢 Accessibility Permission GRANTED by user!")
+                    timer.invalidate()
+                    self?.permissionPollTimer = nil
+                    HotkeyManager.shared.registerHotkeys()
+                    HUDOverlayController.shared.show(state: .success(text: "دسترسی Accessibility تایید شد ✅"))
+                }
+            }
+        }
+        
+        // 2. Microphone permission
         audioRecorder.requestMicrophonePermission { granted in
             if !granted {
                 print("Warning: Microphone permission not granted.")

mac/src/FocusedInputSync.swift

@@ -1,5 +1,6 @@
 import Cocoa
 import ApplicationServices
+import CoreGraphics
 
 public struct MacInputState: Codable {
     public let source: String
@@ -42,7 +43,44 @@ public final class FocusedInputSync {
         AXUIElementSetAttributeValue(systemWideElement, "AXManualAccessibility" as CFString, kCFBooleanTrue)
     }
     
-    private func findFocusedDescendant(_ elem: AXUIElement) -> AXUIElement? {
+    /// Resolves the actual user-facing application, bypassing system overlays
+    public func getRealFrontmostApp() -> NSRunningApplication? {
+        if let front = NSWorkspace.shared.frontmostApplication,
+           front.activationPolicy == .regular,
+           let bundleId = front.bundleIdentifier,
+           !bundleId.contains("notificationcenter"),
+           !bundleId.contains("controlcenter"),
+           !bundleId.contains("WindowManager"),
+           !bundleId.contains("Soniox") {
+            return front
+        }
+        
+        let windowList = CGWindowListCopyWindowInfo([.optionOnScreenOnly, .excludeDesktopElements], kCGNullWindowID) as? [[String: Any]] ?? []
+        for win in windowList {
+            let layer = win[kCGWindowLayer as String] as? Int ?? -1
+            let pid = win[kCGWindowOwnerPID as String] as? pid_t ?? 0
+            let bounds = win[kCGWindowBounds as String] as? [String: Any] ?? [:]
+            let width = bounds["Width"] as? CGFloat ?? 0
+            let height = bounds["Height"] as? CGFloat ?? 0
+            
+            if layer == 0 && width > 100 && height > 100 {
+                if let app = NSRunningApplication(processIdentifier: pid),
+                   app.activationPolicy == .regular,
+                   let bundleId = app.bundleIdentifier,
+                   !bundleId.contains("notificationcenter"),
+                   !bundleId.contains("controlcenter"),
+                   !bundleId.contains("WindowManager"),
+                   !bundleId.contains("Soniox") {
+                    return app
+                }
+            }
+        }
+        return NSWorkspace.shared.frontmostApplication
+    }
+    
+    private func findFocusedDescendant(_ elem: AXUIElement, depth: Int = 0) -> AXUIElement? {
+        if depth > 12 { return nil }
+        
         var isFocusedObj: CFTypeRef?
         if AXUIElementCopyAttributeValue(elem, kAXFocusedAttribute as CFString, &isFocusedObj) == .success,
            let isFocused = isFocusedObj as? Bool, isFocused {
@@ -58,7 +96,7 @@ public final class FocusedInputSync {
         if AXUIElementCopyAttributeValue(elem, kAXChildrenAttribute as CFString, &childrenObj) == .success,
            let children = childrenObj as? [AXUIElement] {
             for child in children {
-                if let found = findFocusedDescendant(child) {
+                if let found = findFocusedDescendant(child, depth: depth + 1) {
                     return found
                 }
             }
@@ -66,8 +104,8 @@ public final class FocusedInputSync {
         return nil
     }
     
-    public func getFocusedElement() -> (AXUIElement, String)? {
-        guard let frontApp = NSWorkspace.shared.frontmostApplication else { return nil }
+    public func getFocusedElement() -> (AXUIElement?, String) {
+        guard let frontApp = getRealFrontmostApp() else { return (nil, "App") }
         let appName = frontApp.localizedName ?? "App"
         let appElem = AXUIElementCreateApplication(frontApp.processIdentifier)
         
@@ -76,7 +114,7 @@ public final class FocusedInputSync {
         
         var targetElem: AXUIElement?
         
-        // 1. Try system wide focused element
+        // 1. System-wide focused element
         var focusedObj: CFTypeRef?
         if AXUIElementCopyAttributeValue(systemWideElement, kAXFocusedUIElementAttribute as CFString, &focusedObj) == .success,
            let obj = focusedObj {
@@ -89,7 +127,7 @@ public final class FocusedInputSync {
             }
         }
         
-        // 2. Try App focused element
+        // 2. App focused element
         if targetElem == nil {
             if AXUIElementCopyAttributeValue(appElem, kAXFocusedUIElementAttribute as CFString, &focusedObj) == .success,
                let obj = focusedObj {
@@ -103,27 +141,53 @@ public final class FocusedInputSync {
             }
         }
         
-        // 3. Recursive search in tree
+        // 3. App focused window element
         if targetElem == nil {
-            targetElem = findFocusedDescendant(appElem)
+            var focusedWinObj: CFTypeRef?
+            if AXUIElementCopyAttributeValue(appElem, kAXFocusedWindowAttribute as CFString, &focusedWinObj) == .success,
+               let win = focusedWinObj {
+                var winFocObj: CFTypeRef?
+                if AXUIElementCopyAttributeValue((win as! AXUIElement), kAXFocusedUIElementAttribute as CFString, &winFocObj) == .success,
+                   let obj = winFocObj {
+                    targetElem = (obj as! AXUIElement)
+                }
+            }
         }
         
-        if let elem = targetElem {
-            return (elem, appName)
+        // 4. Recursive search in tree
+        if targetElem == nil {
+            targetElem = findFocusedDescendant(appElem)
         }
-        return nil
+        
+        return (targetElem, appName)
     }
     
-    /// Inspects the current focused element and returns a state snapshot if changed
+    /// Inspects current focused element on Mac and returns state snapshot if user typed on Mac
     public func inspectCurrentState() -> MacInputState? {
         let now = Date().timeIntervalSince1970
         if isApplyingRemoteChange || now < remoteChangeExpiryTime {
             return nil
         }
         
-        guard let (elem, appName) = getFocusedElement() else { return nil }
+        let (elemOpt, appName) = getFocusedElement()
+        let appChanged = (appName != lastObservedApp)
+        if appChanged {
+            lastObservedApp = appName
+        }
+        
+        guard let elem = elemOpt else {
+            if appChanged && !lastObservedText.isEmpty {
+                localRevision += 1
+                return MacInputState(app: appName, text: lastObservedText, cursor: lastObservedCursor, selection: 0, revision: localRevision)
+            }
+            return nil
+        }
+        
+        var roleObj: CFTypeRef?
+        AXUIElementCopyAttributeValue(elem, kAXRoleAttribute as CFString, &roleObj)
+        let role = roleObj as? String ?? ""
+        let isTextRole = (role == "AXTextField" || role == "AXTextArea" || role == "AXComboBox" || role == "AXSearchField")
         
-        // Extract text
         var text = ""
         var valObj: CFTypeRef?
         if AXUIElementCopyAttributeValue(elem, kAXValueAttribute as CFString, &valObj) == .success,
@@ -135,7 +199,7 @@ public final class FocusedInputSync {
             }
         }
         
-        if text.isEmpty {
+        if text.isEmpty && isTextRole {
             var countObj: CFTypeRef?
             if AXUIElementCopyAttributeValue(elem, kAXNumberOfCharactersAttribute as CFString, &countObj) == .success,
                let count = countObj as? Int, count > 0 {
@@ -150,7 +214,11 @@ public final class FocusedInputSync {
             }
         }
         
-        // Extract cursor & selection
+        // Ghost empty suppression for non-native containers
+        if text.isEmpty && !lastObservedText.isEmpty && !isTextRole {
+            return nil
+        }
+        
         var cursor = text.count
         var selLen = 0
         var rangeObj: CFTypeRef?
@@ -163,8 +231,7 @@ public final class FocusedInputSync {
             }
         }
         
-        // Check if text or cursor actually changed on Mac
-        if text == lastObservedText && cursor == lastObservedCursor && appName == lastObservedApp {
+        if text == lastObservedText && cursor == lastObservedCursor && !appChanged {
             return nil
         }
         
@@ -176,84 +243,141 @@ public final class FocusedInputSync {
         return MacInputState(app: appName, text: text, cursor: cursor, selection: selLen, revision: localRevision)
     }
     
-    /// Applies updated full text or inserts at cursor position directly into active Mac input
+    /// Perfectly mirrors the full text to the active Mac input box in real-time
     @discardableResult
     public func applyRemoteUpdate(text: String, cursor: Int? = nil, isFullReplace: Bool = true) -> Bool {
         isApplyingRemoteChange = true
-        remoteChangeExpiryTime = Date().timeIntervalSince1970 + 0.45 // 450ms quiet window
+        remoteChangeExpiryTime = Date().timeIntervalSince1970 + 0.25
         
         defer {
-            DispatchQueue.main.asyncAfter(deadline: .now() + 0.45) {
+            DispatchQueue.main.asyncAfter(deadline: .now() + 0.25) {
                 self.isApplyingRemoteChange = false
             }
         }
         
-        // 1. Clean and normalize text
-        var cleanText = text.components(separatedBy: .newlines).joined(separator: " ")
-        cleanText = cleanText.replacingOccurrences(of: "\t", with: " ")
-        cleanText = cleanText.replacingOccurrences(of: "\\s+", with: " ", options: .regularExpression)
+        // Normalize line endings and preserve intentional multiline text (\n)
+        var cleanText = text.replacingOccurrences(of: "\r\n", with: "\n").replacingOccurrences(of: "\r", with: "\n")
         cleanText = cleanText.trimmingCharacters(in: .whitespacesAndNewlines)
         
         let targetCursor = cursor ?? cleanText.count
         lastObservedText = cleanText
         lastObservedCursor = targetCursor
         
-        guard let (elem, appName) = getFocusedElement() else {
-            // Fallback: clipboard paste
-            lastObservedApp = NSWorkspace.shared.frontmostApplication?.localizedName ?? "App"
-            return pasteViaKeystroke(cleanText, isFullReplace: isFullReplace)
-        }
+        let (_, appName) = getFocusedElement()
         lastObservedApp = appName
         
+        // Universal Quartz CGEvent Keystroke Engine (Cmd+A -> Cmd+V / Backspace or pure Cmd+V)
+        // Works 100% reliably across native, web, and Electron/Chromium apps (e.g. Antigravity, VS Code, Slack, Firefox)
+        print("FocusedInputSync: 🚀 Injecting text into '\(appName)' (replace: \(isFullReplace), len: \(cleanText.count), lines: \(cleanText.components(separatedBy: "\n").count))")
         if isFullReplace {
-            // Try setting AXValue directly
-            let setErr = AXUIElementSetAttributeValue(elem, kAXValueAttribute as CFString, cleanText as CFTypeRef)
-            if setErr == .success {
-                if let cursor = cursor {
-                    var range = CFRange(location: min(cursor, cleanText.count), length: 0)
-                    if let axRange = AXValueCreate(.cfRange, &range) {
-                        AXUIElementSetAttributeValue(elem, kAXSelectedTextRangeAttribute as CFString, axRange)
-                    }
-                }
-                print("FocusedInputSync: ✅ Updated AXValue for \(appName)")
-                return true
-            }
+            return executeCleanFullReplace(cleanText)
         } else {
-            // Try setting selected text
-            let setSelErr = AXUIElementSetAttributeValue(elem, kAXSelectedTextAttribute as CFString, cleanText as CFTypeRef)
-            if setSelErr == .success {
-                print("FocusedInputSync: ✅ Inserted text via AXSelectedText for \(appName)")
-                return true
-            }
+            return pasteOnlyViaCleanKeystroke(cleanText)
         }
-        
-        return pasteViaKeystroke(cleanText, isFullReplace: isFullReplace)
     }
     
-    private func pasteViaKeystroke(_ text: String, isFullReplace: Bool) -> Bool {
+    /**
+     * Inserts speech directly at active cursor via clean Cmd+V
+     */
+    private func pasteOnlyViaCleanKeystroke(_ text: String) -> Bool {
+        guard !text.isEmpty else { return true }
+        
         let pasteboard = NSPasteboard.general
         pasteboard.clearContents()
         pasteboard.setString(text, forType: .string)
         
         let src = CGEventSource(stateID: .hidSystemState)
+        let kVK_ANSI_V: CGKeyCode = 9
         
-        if isFullReplace {
+        // Paste: Cmd + V
+        if let vDown = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_V, keyDown: true) {
+            vDown.flags = .maskCommand
+            vDown.post(tap: .cghidEventTap)
+            vDown.post(tap: .cgSessionEventTap)
+        }
+        usleep(12000)
+        
+        if let vUp = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_V, keyDown: false) {
+            vUp.flags = .maskCommand
+            vUp.post(tap: .cghidEventTap)
+            vUp.post(tap: .cgSessionEventTap)
+        }
+        usleep(10000)
+        return true
+    }
+    
+    /**
+     * Replaces the entire content of active input box cleanly in real-time.
+     * If text is empty: Cmd+A -> Backspace.
+     * If text is non-empty: Cmd+A -> Cmd+V.
+     */
+    private func executeCleanFullReplace(_ text: String) -> Bool {
+        let src = CGEventSource(stateID: .hidSystemState)
+        let kVK_ANSI_A: CGKeyCode = 0
+        let kVK_ANSI_V: CGKeyCode = 9
+        let kVK_Delete: CGKeyCode = 51
+        
+        if text.isEmpty {
             // Select all: Cmd + A
-            let aDown = CGEvent(keyboardEventSource: src, virtualKey: 0, keyDown: true)
-            aDown?.flags = .maskCommand
-            let aUp = CGEvent(keyboardEventSource: src, virtualKey: 0, keyDown: false)
-            aDown?.post(tap: .cghidEventTap)
-            aUp?.post(tap: .cghidEventTap)
+            if let aDown = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_A, keyDown: true) {
+                aDown.flags = .maskCommand
+                aDown.post(tap: .cghidEventTap)
+                aDown.post(tap: .cgSessionEventTap)
+            }
+            usleep(8000)
+            if let aUp = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_A, keyDown: false) {
+                aUp.flags = .maskCommand
+                aUp.post(tap: .cghidEventTap)
+                aUp.post(tap: .cgSessionEventTap)
+            }
+            usleep(15000)
             
-            usleep(20000)
+            // Backspace to clear
+            if let delDown = CGEvent(keyboardEventSource: src, virtualKey: kVK_Delete, keyDown: true) {
+                delDown.post(tap: .cghidEventTap)
+                delDown.post(tap: .cgSessionEventTap)
+            }
+            usleep(8000)
+            if let delUp = CGEvent(keyboardEventSource: src, virtualKey: kVK_Delete, keyDown: false) {
+                delUp.post(tap: .cghidEventTap)
+                delUp.post(tap: .cgSessionEventTap)
+            }
+            return true
         }
         
-        // Paste: Cmd + V
-        let vDown = CGEvent(keyboardEventSource: src, virtualKey: 9, keyDown: true)
-        vDown?.flags = .maskCommand
-        let vUp = CGEvent(keyboardEventSource: src, virtualKey: 9, keyDown: false)
-        vDown?.post(tap: .cghidEventTap)
-        vUp?.post(tap: .cghidEventTap)
+        // Set clipboard
+        let pasteboard = NSPasteboard.general
+        pasteboard.clearContents()
+        pasteboard.setString(text, forType: .string)
+        
+        // 1. Select all: Cmd + A
+        if let aDown = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_A, keyDown: true) {
+            aDown.flags = .maskCommand
+            aDown.post(tap: .cghidEventTap)
+            aDown.post(tap: .cgSessionEventTap)
+        }
+        usleep(8000)
+        
+        if let aUp = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_A, keyDown: false) {
+            aUp.flags = .maskCommand
+            aUp.post(tap: .cghidEventTap)
+            aUp.post(tap: .cgSessionEventTap)
+        }
+        usleep(18000) // 18ms for selection to settle
+        
+        // 2. Paste: Cmd + V
+        if let vDown = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_V, keyDown: true) {
+            vDown.flags = .maskCommand
+            vDown.post(tap: .cghidEventTap)
+            vDown.post(tap: .cgSessionEventTap)
+        }
+        usleep(12000)
+        
+        if let vUp = CGEvent(keyboardEventSource: src, virtualKey: kVK_ANSI_V, keyDown: false) {
+            vUp.flags = .maskCommand
+            vUp.post(tap: .cghidEventTap)
+            vUp.post(tap: .cgSessionEventTap)
+        }
         
         return true
     }

mac/src/HUDOverlay.swift

@@ -8,10 +8,16 @@ public enum HUDState {
     case error(message: String)
 }
 
+final class NonActivatingFloatingPanel: NSPanel {
+    override var canBecomeKey: Bool { return false }
+    override var canBecomeMain: Bool { return false }
+    override var acceptsFirstResponder: Bool { return false }
+}
+
 public final class HUDOverlayController {
     public static let shared = HUDOverlayController()
     
-    private var window: NSPanel?
+    private var window: NonActivatingFloatingPanel?
     private var visualEffectView: NSVisualEffectView?
     private var iconImageView: NSImageView?
     private var titleLabel: NSTextField?
@@ -27,7 +33,7 @@ public final class HUDOverlayController {
         let width: CGFloat = 460
         let height: CGFloat = 80
         
-        let panel = NSPanel(
+        let panel = NonActivatingFloatingPanel(
             contentRect: NSRect(x: 0, y: 0, width: width, height: height),
             styleMask: [.borderless, .nonactivatingPanel],
             backing: .buffered,
@@ -39,6 +45,7 @@ public final class HUDOverlayController {
         panel.backgroundColor = .clear
         panel.hasShadow = true
         panel.ignoresMouseEvents = true
+        panel.becomesKeyOnlyIfNeeded = false
         panel.collectionBehavior = [.canJoinAllSpaces, .fullScreenAuxiliary]
         
         let visualEffect = NSVisualEffectView(frame: NSRect(x: 0, y: 0, width: width, height: height))

mac/src/RelayClient.swift

@@ -19,6 +19,9 @@ public final class RelayClient: NSObject, URLSessionWebSocketDelegate {
     
     public var onRemoteUpdateReceived: ((String, Int?, Bool) -> Void)?
     
+    private var lastInjectedText: String = ""
+    private var lastInjectedTime: Double = 0
+    
     override private init() {
         super.init()
         let config = URLSessionConfiguration.default
@@ -97,7 +100,7 @@ public final class RelayClient: NSObject, URLSessionWebSocketDelegate {
             return
         }
         
-        let action = json["action"] as? String ?? json["type"] as? String
+        let action = json["action"] as? String ?? json["type"] as? String ?? ""
         let source = json["source"] as? String ?? ""
         
         // Ignore echo messages originated by Mac itself
@@ -106,9 +109,18 @@ public final class RelayClient: NSObject, URLSessionWebSocketDelegate {
         let updateText = json["text"] as? String ?? json["insert_text"] as? String ?? ""
         let cursor = json["cursor"] as? Int ?? json["cursor_pos"] as? Int
         
-        if action == "paste" || action == "set_text" || action == "update_input" || action == "phone_input_edit" || action == "sync_state" {
-            let isFullReplace = (action != "insert_at_cursor")
-            print("RelayClient: ⚡ Received remote Android update: '\(updateText.prefix(30))...' (replace: \(isFullReplace), cursor: \(cursor ?? -1))")
+        if action == "sync_state" || action == "phone_input_edit" || action == "update_input" || action == "force_replace" || action == "insert_speech" || action == "paste" {
+            let now = Date().timeIntervalSince1970
+            
+            // Drop rapid duplicate transmissions within 150ms (unless force_replace)
+            if updateText == lastInjectedText && (now - lastInjectedTime) < 0.15 && action != "force_replace" {
+                return
+            }
+            lastInjectedText = updateText
+            lastInjectedTime = now
+            
+            let isFullReplace = (action != "insert_speech" && action != "speech_insert")
+            print("RelayClient: ⚡ Applying phone edit: '\(updateText.prefix(30))...' (replace: \(isFullReplace), len: \(updateText.count))")
             DispatchQueue.main.async {
                 self.onRemoteUpdateReceived?(updateText, cursor, isFullReplace)
             }

server/relay_server.py

@@ -1,9 +1,9 @@
 #!/usr/bin/env python3
 """
-Soniox Collaborative Bi-Directional Input Synchronization Gateway (v5.0)
-- Real-time Google Docs / Figma style collaborative mirroring between Mac and Android.
-- Monotonic revision counters, echo-loop suppression, and sub-15ms WebSocket routing.
-- Cursor-aware voice insertion and instant text editing.
+Soniox Collaborative Bi-Directional Input Synchronization Gateway (v5.4)
+- Ultra-low latency streaming speech recognition with pre-warmed Soniox pool.
+- Strict artifact & lone-punctuation suppression (e.g. «, », ., quotes).
+- Clean single-injection pipeline on speech finalization.
 """
 
 import asyncio
@@ -59,7 +59,7 @@ def is_ws_open(ws) -> bool:
 
 class SonioxPool:
     """Pre-warms upstream WebSockets to Soniox for 0ms speech start delay."""
-    def __init__(self, size=3):
+    def __init__(self, size=4):
         self._pool = asyncio.Queue(maxsize=size)
         self._refilling = False
 
@@ -83,7 +83,7 @@ class SonioxPool:
                 SONIOX_WS_URL,
                 additional_headers=SONIOX_HEADERS,
                 open_timeout=3.5,
-                ping_interval=20,
+                ping_interval=15,
             )
         except Exception as e:
             logger.error("Failed to connect to upstream Soniox: %s", e)
@@ -105,12 +105,25 @@ class SonioxPool:
 
 soniox_pool = SonioxPool()
 
-def sanitize_and_flatten_text(text: str) -> str:
+def sanitize_text(text: str, allow_multiline: bool = True) -> str:
     if not text:
         return ""
-    flattened = re.sub(r"[\r\n\t]+", " ", text)
-    flattened = re.sub(r"\s+", " ", flattened).strip()
-    return flattened
+    if allow_multiline:
+        normalized = text.replace("\r\n", "\n").replace("\r", "\n")
+        lines = [re.sub(r"[ \t]+", " ", line) for line in normalized.split("\n")]
+        cleaned = "\n".join(lines).strip("\r\n")
+    else:
+        cleaned = re.sub(r"[\r\n\t]+", " ", text)
+        cleaned = re.sub(r"\s+", " ", cleaned).strip()
+    
+    # Suppress lone punctuation artifacts (e.g. «, », ., ,, !, ?, etc.)
+    if re.fullmatch(r"[\s«»\.\,\،\؛\؟\!\?\:\;\-\–—\"\'\(\)\[\]\{\}]+", cleaned):
+        return ""
+        
+    return cleaned
+
+def sanitize_and_flatten_text(text: str) -> str:
+    return sanitize_text(text, allow_multiline=False)
 
 async def broadcast_state(payload_dict: dict, exclude_ws=None):
     """Broadcasts state snapshot to all connected clients (Mac and Phone) except sender."""
@@ -124,6 +137,10 @@ async def broadcast_state(payload_dict: dict, exclude_ws=None):
         current_room_state.update(payload_dict)
         payload_str = json.dumps(payload_dict, ensure_ascii=False)
 
+    logger.info("📡 Broadcasting %s (len: %d) to %d Macs, %d Phones", 
+                payload_dict.get("type"), len(payload_dict.get("text", "")), 
+                len(connected_mac_websockets), len(connected_phone_websockets))
+
     # 1. Send to Phone clients
     dead_phones = set()
     for ws in list(connected_phone_websockets):
@@ -156,7 +173,7 @@ async def broadcast_state(payload_dict: dict, exclude_ws=None):
 
 async def handle_phone_stream_ws(request):
     """Duplex WebSocket for Android client (Real-time Live Sync + Audio Dictation)."""
-    ws = web.WebSocketResponse(heartbeat=12.0)
+    ws = web.WebSocketResponse(heartbeat=10.0, autoping=True)
     await ws.prepare(request)
     client_ip = request.remote
     logger.info("📱 Android Client Connected: %s", client_ip)
@@ -238,10 +255,15 @@ async def handle_phone_stream_ws(request):
                 msg_type = data.get("type") or data.get("action")
                 sid = data.get("session_id", f"sess_{int(time.time()*1000)}")
 
-                if msg_type == "sync_state" or msg_type == "phone_input_edit" or msg_type == "update_input":
-                    # Phone edited text: broadcast to Mac immediately!
+                # ALL text / sync / insert operations must be broadcast to Mac!
+                if msg_type in ("sync_state", "insert_speech", "speech_insert", "phone_input_edit", "update_input", "paste"):
+                    is_speech = msg_type in ("insert_speech", "speech_insert")
+                    clean_text = sanitize_text(data.get("text", ""), allow_multiline=(not is_speech))
+                    if not clean_text and is_speech:
+                        continue  # Do not broadcast empty speech or lone quotes
+                    data["text"] = clean_text
                     data["source"] = "android"
-                    data["type"] = "sync_state"
+                    data["type"] = msg_type
                     await broadcast_state(data, exclude_ws=ws)
 
                 elif msg_type == "start":
@@ -264,7 +286,7 @@ async def handle_phone_stream_ws(request):
                     if active_soniox_ws and is_ws_open(active_soniox_ws):
                         await active_soniox_ws.send(json.dumps({"type": "finalize"}))
                         try:
-                            await asyncio.wait_for(stop_event.wait(), timeout=0.55)
+                            await asyncio.wait_for(stop_event.wait(), timeout=0.35)
                         except asyncio.TimeoutError:
                             pass
 
@@ -311,7 +333,7 @@ async def handle_phone_stream_ws(request):
 
 async def handle_mac_ws(request):
     """Persistent WebSocket for Mac Bridge (receives live Mac typing & pushes edits)."""
-    ws = web.WebSocketResponse(heartbeat=10.0)
+    ws = web.WebSocketResponse(heartbeat=10.0, autoping=True)
     await ws.prepare(request)
     client_ip = request.remote
     logger.info("🖥️ Mac client connected: %s", client_ip)
@@ -348,7 +370,7 @@ async def handle_health(request):
         state_copy = dict(current_room_state)
     return web.json_response({
         "status": "ok",
-        "service": "Soniox Collaborative Sync Gateway v5.0",
+        "service": "Soniox Collaborative Sync Gateway v5.4",
         "connected_macs": len(connected_mac_websockets),
         "connected_phones": len(connected_phone_websockets),
         "current_app": state_copy.get("app", ""),
@@ -362,7 +384,7 @@ async def handle_paste(request):
         text = data.get("text", "")
         cursor = data.get("cursor_pos") or data.get("cursor")
         data["source"] = "http_post"
-        data["type"] = "sync_state"
+        data["type"] = "update_input"
         await broadcast_state(data)
         return web.json_response({"status": "synced", "revision": current_room_state["revision"]})
     except Exception as e: