feat(cot): DeepSeek-R1 CoT accordion, multimodal EXIF downsampling, TokenStorage hardening & test suite

.github/workflows/android.yml

@@ -0,0 +1,68 @@
+name: Android CI & Validation
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build-and-test:
+    name: Build, Test & Lint
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+
+    steps:
+      - name: Checkout Repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Set up JDK 17
+        uses: actions/setup-java@v4
+        with:
+          distribution: 'temurin'
+          java-version: '17'
+
+      - name: Setup Gradle
+        uses: gradle/actions/setup-gradle@v4
+
+      - name: Grant execute permission for gradlew
+        run: chmod +x gradlew
+
+      - name: Run Unit Tests
+        run: ./gradlew testDebugUnitTest --no-daemon --stacktrace
+
+      - name: Run Android Lint
+        run: ./gradlew lintDebug --no-daemon
+
+      - name: Assemble Debug APK
+        run: ./gradlew assembleDebug --no-daemon
+
+      - name: Upload Debug APK
+        uses: actions/upload-artifact@v4
+        with:
+          name: aura-local-ai-debug-apk
+          path: app/build/outputs/apk/debug/app-debug.apk
+          if-no-files-found: error
+          retention-days: 14
+
+      - name: Upload Unit Test Reports
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: unit-test-reports
+          path: app/build/reports/tests/testDebugUnitTest/
+          retention-days: 7
+
+      - name: Upload Lint Report
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: lint-reports
+          path: app/build/reports/lint-results-debug.html
+          retention-days: 7

README.md

@@ -44,7 +44,7 @@ The app includes built-in presets for several highly-capable, lightweight models
 | --- | --- |
 | **DeepSeek-R1 Distill Qwen 1.5B**<br/>• Parameters: 1.5B \| Size: ~1.7 GB<br/>• Min. RAM: 6 GB+ (Offline Reasoning) | **Qwen 2.5 Coder 3B Instruct**<br/>• Parameters: 3B \| Size: ~2.9 GB<br/>• Min. RAM: 8 GB+ (Coding Expert) |
 | **Qwen 2.5 1.5B Instruct**<br/>• Parameters: 1.5B \| Size: ~1.5 GB<br/>• Min. RAM: 6 GB+ (General Knowledge) | **Google Gemma 4 E2B Instruct**<br/>• Parameters: 2B \| Size: ~2.4 GB<br/>• Min. RAM: 6 GB+ (Multimodal Vision) |
-| **Qwen 3 4B**<br/>• Parameters: 4B \| Size: ~2.5 GB<br/>• Min. RAM: 8 GB+ (High Performance) | **Google Gemma 4 E4B Instruct**<br/>• Parameters: 4B \| Size: ~3.4 GB<br/>• Min. RAM: 8 GB+ (High-Res Multimodal) |
+| **Qwen 3 4B**<br/>• Parameters: 4B \| Size: ~2.5 GB<br/>• Min. RAM: 8 GB+ (High Performance) | **Google Gemma 4 E4B Instruct**<br/>• Parameters: 4B \| Size: ~3.4 GB<br/>• Min. RAM: 12 GB+ (High-Res Multimodal) |
 | **Qwen 2.5 0.5B Instruct**<br/>• Parameters: 0.5B \| Size: ~0.5 GB<br/>• Min. RAM: 4 GB+ (Ultra-Fast) | |
 
 ---

app/build.gradle.kts

@@ -41,6 +41,11 @@ android {
       shaders = false
     }
 
+    lint {
+      abortOnError = false
+      checkReleaseBuilds = false
+    }
+
 
 
     packaging {

app/src/main/java/com/example/auralocalai/data/LlmInferenceEngine.kt

@@ -177,7 +177,7 @@ class LlmInferenceEngine(private val context: Context) {
                         }
                         gpuError = ramError
                     } else {
-                        val gpuStage = if (npuError != null) "NPU unavailable — initializing GPU backend…" else "Initializing GPU backend…"
+                        val gpuStage = if (npuError != null) "NPU unavailable — compiling GPU shaders…" else "Compiling GPU shaders…"
                         onStageUpdate?.invoke(gpuStage)
                         try {
                             val config = EngineConfig(
@@ -186,14 +186,25 @@ class LlmInferenceEngine(private val context: Context) {
                                 cacheDir = context.cacheDir.absolutePath
                             )
                             val newEngine = Engine(config)
-                            newEngine.initialize()
+                            if (preferredBackend == "AUTO") {
+                                val initOk = kotlinx.coroutines.withTimeoutOrNull(40_000L) {
+                                    newEngine.initialize()
+                                    true
+                                }
+                                if (initOk == null) {
+                                    throw RuntimeException("GPU shader compilation timed out after 40s")
+                                }
+                            } else {
+                                newEngine.initialize()
+                            }
                             engine = newEngine
                             conversation = newEngine.createConversation()
                             currentModelPath = modelPath
                             activeBackend = "GPU"
                             loaded = true
                         } catch (e: Throwable) {
                             gpuError = e
+                            android.util.Log.w("LlmInferenceEngine", "GPU initialization failed or timed out: ${e.message}")
                             // If CPU fallback is NOT allowed or GPU_ONLY is preferred, fail immediately
                             if (restriction != LlmBackendRestriction.ANY || preferredBackend == "GPU_ONLY") {
                                 val msg = buildString {
@@ -203,7 +214,7 @@ class LlmInferenceEngine(private val context: Context) {
                                 }
                                 return@withContext Result.failure(Exception(msg, e))
                             }
-                    }
+                        }
                 }
             }
             }

app/src/main/java/com/example/auralocalai/data/ModelDownloadService.kt

@@ -9,15 +9,19 @@ import android.content.Intent
 import android.content.pm.ServiceInfo
 import android.os.Build
 import android.os.IBinder
+import android.util.Log
 import androidx.core.app.NotificationCompat
 import com.example.auralocalai.MainActivity
+import com.example.auralocalai.R
 import kotlinx.coroutines.CoroutineScope
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.Job
 import kotlinx.coroutines.flow.MutableStateFlow
 import kotlinx.coroutines.launch
 import java.io.File
 
+private const val TAG = "ModelDownloadService"
+
 sealed interface ServiceDownloadState {
     data object Idle : ServiceDownloadState
     data class Progress(
@@ -38,7 +42,6 @@ class ModelDownloadService : Service() {
     private val serviceJob = Job()
     private val serviceScope = CoroutineScope(Dispatchers.IO + serviceJob)
     private var activeDownloadJob: Job? = null
-
     private lateinit var downloader: ModelDownloader
     private lateinit var notificationManager: NotificationManager
 
@@ -66,7 +69,7 @@ class ModelDownloadService : Service() {
             return START_NOT_STICKY
         }
 
-        // Start Foreground Service
+        // Start Foreground Service safely
         startForegroundServiceCompat(modelId, fileName)
 
         // Cancel any active download before starting a new one
@@ -130,15 +133,14 @@ class ModelDownloadService : Service() {
             }
         }
 
-        return START_STICKY
+        return START_NOT_STICKY
     }
 
     override fun onBind(intent: Intent?): IBinder? = null
 
     override fun onDestroy() {
         activeDownloadJob?.cancel()
         serviceJob.cancel()
-        downloadState.value = ServiceDownloadState.Idle
         super.onDestroy()
     }
 
@@ -150,66 +152,79 @@ class ModelDownloadService : Service() {
                 NotificationManager.IMPORTANCE_LOW
             ).apply {
                 description = "Shows progress of model downloads running in the background"
+                setShowBadge(false)
             }
             notificationManager.createNotificationChannel(channel)
         }
     }
 
     private fun startForegroundServiceCompat(modelId: String, fileName: String) {
-        val notification = BuildNotification(
-            title = "Downloading Model",
-            content = "Starting download for $fileName...",
-            progress = 0,
-            indeterminate = true
-        )
-
-        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.Q) {
-            startForeground(
-                NOTIFICATION_ID,
-                notification,
-                ServiceInfo.FOREGROUND_SERVICE_TYPE_DATA_SYNC
+        try {
+            val notification = buildNotification(
+                title = "Downloading Model",
+                content = "Starting download for $fileName...",
+                progress = 0,
+                indeterminate = true
             )
-        } else {
-            startForeground(NOTIFICATION_ID, notification)
+
+            if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.Q) {
+                startForeground(
+                    NOTIFICATION_ID,
+                    notification,
+                    ServiceInfo.FOREGROUND_SERVICE_TYPE_DATA_SYNC
+                )
+            } else {
+                startForeground(NOTIFICATION_ID, notification)
+            }
+        } catch (e: Exception) {
+            Log.w(TAG, "startForeground failed (ignoring to allow download to proceed): ${e.message}")
         }
     }
 
     private fun updateProgressNotification(modelId: String, fileName: String, percentage: Int, speed: Double) {
-        val speedText = formatSpeed(speed)
-        val notification = BuildNotification(
-            title = "Downloading $fileName",
-            content = "$percentage% completed • $speedText",
-            progress = percentage,
-            indeterminate = false
-        )
-        notificationManager.notify(NOTIFICATION_ID, notification)
+        try {
+            val speedText = formatSpeed(speed)
+            val notification = buildNotification(
+                title = "Downloading $fileName",
+                content = "$percentage% completed • $speedText",
+                progress = percentage,
+                indeterminate = false
+            )
+            notificationManager.notify(NOTIFICATION_ID, notification)
+        } catch (e: Exception) {
+            Log.d(TAG, "Notification update skipped: ${e.message}")
+        }
     }
 
     private fun showCompletionNotification(modelId: String, fileName: String, success: Boolean) {
-        val title = if (success) "Download Successful" else "Download Failed"
-        val content = if (success) "Successfully downloaded $fileName." else "Failed to download $fileName."
-        val intent = Intent(this, MainActivity::class.java).apply {
-            flags = Intent.FLAG_ACTIVITY_NEW_TASK or Intent.FLAG_ACTIVITY_CLEAR_TASK
-        }
-        val pendingIntent = PendingIntent.getActivity(
-            this,
-            0,
-            intent,
-            PendingIntent.FLAG_UPDATE_CURRENT or PendingIntent.FLAG_IMMUTABLE
-        )
+        try {
+            val title = if (success) "Download Successful" else "Download Failed"
+            val content = if (success) "Successfully downloaded $fileName." else "Failed to download $fileName."
+            val intent = Intent(this, MainActivity::class.java).apply {
+                flags = Intent.FLAG_ACTIVITY_NEW_TASK or Intent.FLAG_ACTIVITY_CLEAR_TASK
+            }
+            val pendingIntent = PendingIntent.getActivity(
+                this,
+                0,
+                intent,
+                PendingIntent.FLAG_UPDATE_CURRENT or PendingIntent.FLAG_IMMUTABLE
+            )
 
-        val notification = NotificationCompat.Builder(this, CHANNEL_ID)
-            .setSmallIcon(android.R.drawable.stat_sys_download_done)
-            .setContentTitle(title)
-            .setContentText(content)
-            .setContentIntent(pendingIntent)
-            .setAutoCancel(true)
-            .build()
+            val notification = NotificationCompat.Builder(this, CHANNEL_ID)
+                .setSmallIcon(if (success) R.drawable.ic_download_done else R.drawable.ic_download)
+                .setContentTitle(title)
+                .setContentText(content)
+                .setContentIntent(pendingIntent)
+                .setAutoCancel(true)
+                .build()
 
-        notificationManager.notify(NOTIFICATION_ID + 1, notification)
+            notificationManager.notify(NOTIFICATION_ID + 1, notification)
+        } catch (e: Exception) {
+            Log.d(TAG, "Completion notification skipped: ${e.message}")
+        }
     }
 
-    private fun BuildNotification(title: String, content: String, progress: Int, indeterminate: Boolean): android.app.Notification {
+    private fun buildNotification(title: String, content: String, progress: Int, indeterminate: Boolean): android.app.Notification {
         val intent = Intent(this, MainActivity::class.java).apply {
             flags = Intent.FLAG_ACTIVITY_NEW_TASK or Intent.FLAG_ACTIVITY_CLEAR_TASK
         }
@@ -221,21 +236,22 @@ class ModelDownloadService : Service() {
         )
 
         return NotificationCompat.Builder(this, CHANNEL_ID)
-            .setSmallIcon(android.R.drawable.stat_sys_download)
+            .setSmallIcon(R.drawable.ic_download)
             .setContentTitle(title)
             .setContentText(content)
             .setProgress(100, progress, indeterminate)
             .setContentIntent(pendingIntent)
             .setOngoing(true)
+            .setSilent(true)
             .build()
     }
 
     private fun formatSpeed(bytesPerSec: Double): String {
         val mbps = bytesPerSec / (1024 * 1024)
         if (mbps >= 1.0) {
-            return String.format("%.1f MB/s", mbps)
+            return String.format(java.util.Locale.US, "%.1f MB/s", mbps)
         }
         val kbps = bytesPerSec / 1024
-        return String.format("%.1f KB/s", kbps)
+        return String.format(java.util.Locale.US, "%.1f KB/s", kbps)
     }
 }

app/src/main/java/com/example/auralocalai/data/ModelPreset.kt

@@ -23,9 +23,23 @@ data class ModelPreset(
     val backendRestriction: LlmBackendRestriction = LlmBackendRestriction.ANY,
     val quantization: String = "INT4",
     val parameterCount: String = "Unknown",
-    val contextLength: String = "4,096 tokens"
+    val contextLength: String = "4,096 tokens",
+    val isReasoningModel: Boolean = false
 ) {
     companion object {
+        fun isReasoningModel(modelId: String?): Boolean {
+            if (modelId == null) return false
+            val preset = presets.find { it.id.equals(modelId, ignoreCase = true) }
+            if (preset != null) return preset.isReasoningModel
+            return modelId.contains("deepseek", ignoreCase = true) || modelId.contains("qwq", ignoreCase = true)
+        }
+
+        fun isKnownNonReasoningModel(modelId: String?): Boolean {
+            if (modelId == null) return false
+            val preset = presets.find { it.id.equals(modelId, ignoreCase = true) }
+            return preset != null && !preset.isReasoningModel
+        }
+
         val presets = listOf(
             ModelPreset(
                 id = "deepseek-1.5b",
@@ -40,7 +54,8 @@ data class ModelPreset(
                 backendRestriction = LlmBackendRestriction.ANY,
                 quantization = "Q8 (8-bit)",
                 parameterCount = "1.5B",
-                contextLength = "4,096 tokens"
+                contextLength = "4,096 tokens",
+                isReasoningModel = true
             ),
             ModelPreset(
                 id = "qwen-1.5b",
@@ -107,7 +122,7 @@ data class ModelPreset(
                 name = "Google Gemma 4 E4B Instruct (Multimodal)",
                 description = "Google's powerful on-device LLM with 4B parameters. Superior reasoning, math, and coding over E2B with native multimodal vision support (High-Res Multimodal).",
                 sizeLabel = "3.4 GB",
-                ramRequirement = "8 GB+ RAM",
+                ramRequirement = "12 GB+ RAM",
                 downloadUrl = "https://huggingface.co/litert-community/gemma-4-E4B-it-litert-lm/resolve/main/gemma-4-E4B-it.litertlm",
                 fileName = "gemma4-e4b.litertlm",
                 requiresHfToken = false,