Sync desktop/ from omi-desktop (474e1e1f)

desktop/.gitignore

@@ -27,10 +27,6 @@ DerivedData/
 .swiftpm/
 Package.resolved
 
-# Agent bridge build artifacts
-agent-bridge/node_modules/
-agent-bridge/dist/
-
 # Xcode
 *.pbxuser
 *.mode1v3

desktop/CHANGELOG.json

@@ -1,5 +1,36 @@
 {
   "releases": [
+    {
+      "version": "0.10.4",
+      "date": "2026-02-21",
+      "changes": [
+        "Fixed staging channel Option+click gesture being immediately overwritten by beta",
+        "Verification script no longer kills running Dev and Production app instances"
+      ]
+    },
+    {
+      "version": "0.10.3",
+      "date": "2026-02-21",
+      "changes": [
+        "ACP bridge bundled in release builds: AI chat now works reliably without external dependencies",
+        "Update channel label shown in Settings so you can see if you're on Beta or Staging",
+        "Native binaries in ACP bridge node_modules are now properly code-signed"
+      ]
+    },
+    {
+      "version": "0.10.2",
+      "date": "2026-02-20",
+      "changes": [
+        "Recurring task scheduler: AI automatically investigates recurring tasks when they're due",
+        "Parallel task agents: multiple tasks can be investigated by AI simultaneously",
+        "Claude OAuth browser login: sign in to your Claude account via browser instead of token paste",
+        "Separate screen capture and audio recording toggles in Settings for independent control",
+        "Improved task drag-and-drop with cross-category moves and visual drop indicators",
+        "Unified AI bridge: removed legacy Agent SDK, all chat uses the faster ACP bridge",
+        "Session pre-warming for both Opus and Sonnet models for faster first responses",
+        "Fixed stale message desync that could cause AI chat to hang after interruptions"
+      ]
+    },
     {
       "version": "0.10.1",
       "date": "2026-02-20",

desktop/Desktop/Sources/AgentSyncService.swift

@@ -28,6 +28,7 @@ actor AgentSyncService {
     // MARK: - State
 
     private var cursors: [String: SyncCursor] = [:]
+    private var cachedTableColumns: [String: [String]] = [:]
     private var vmIP: String?
     private var authToken: String?
     private var isRunning = false
@@ -223,16 +224,30 @@ actor AgentSyncService {
 
         let cursor = cursors[spec.name] ?? SyncCursor(lastId: 0, lastUpdatedAt: "1970-01-01T00:00:00")
 
-        do {
-            let rows: [[String: Any]] = try await dbPool.read { db in
-                // Get actual column names from the table
-                let columnInfos = try Row.fetchAll(db, sql: "PRAGMA table_info('\(spec.name)')")
-                let allColumns = columnInfos.compactMap { $0["name"] as? String }
-                let columns = allColumns.filter { !spec.excludedColumns.contains($0) }
+        // Resolve columns once and cache — PRAGMA table_info is static at runtime
+        let columns: [String]
+        if let cached = cachedTableColumns[spec.name] {
+            columns = cached
+        } else {
+            do {
+                let fetched: [String] = try await dbPool.read { db in
+                    let columnInfos = try Row.fetchAll(db, sql: "PRAGMA table_info('\(spec.name)')")
+                    let allColumns = columnInfos.compactMap { $0["name"] as? String }
+                    return allColumns.filter { !spec.excludedColumns.contains($0) }
+                }
+                cachedTableColumns[spec.name] = fetched
+                columns = fetched
+            } catch {
+                log("AgentSync: error fetching schema for \(spec.name) — \(error.localizedDescription)")
+                return 0
+            }
+        }
 
-                guard !columns.isEmpty else { return [] }
+        guard !columns.isEmpty else { return 0 }
 
-                let selectCols = columns.map { "\"\($0)\"" }.joined(separator: ", ")
+        do {
+            let selectCols = columns.map { "\"\($0)\"" }.joined(separator: ", ")
+            let rows: [[String: Any]] = try await dbPool.read { db in
                 let sql: String
                 let args: [any DatabaseValueConvertible]
 
@@ -335,6 +350,10 @@ actor AgentSyncService {
 
             if httpResponse.statusCode == 200 {
                 return .success
+            } else if httpResponse.statusCode >= 500 {
+                let body = String(data: data, encoding: .utf8) ?? ""
+                log("AgentSync: push \(table) failed — HTTP \(httpResponse.statusCode): \(body)")
+                return .networkError  // 5xx = server not ready, trigger backoff
             } else {
                 let body = String(data: data, encoding: .utf8) ?? ""
                 log("AgentSync: push \(table) failed — HTTP \(httpResponse.statusCode): \(body)")

desktop/Desktop/Sources/AnalyticsManager.swift

@@ -698,6 +698,11 @@ class AnalyticsManager {
         PostHogManager.shared.tierChanged(tier: tier, reason: reason)
     }
 
+    func chatBridgeModeChanged(from oldMode: String, to newMode: String) {
+        MixpanelManager.shared.chatBridgeModeChanged(from: oldMode, to: newMode)
+        PostHogManager.shared.chatBridgeModeChanged(from: oldMode, to: newMode)
+    }
+
     // MARK: - Settings State
 
     /// Track the current state of key settings (screenshots, memory extraction, notifications)
@@ -802,6 +807,9 @@ class AnalyticsManager {
         props["rewind_retention_days"] = ud.object(forKey: "rewindRetentionDays") as? Double ?? 7.0
         props["rewind_capture_interval"] = ud.object(forKey: "rewindCaptureInterval") as? Double ?? 1.0
 
+        // -- AI Chat Mode --
+        props["chat_bridge_mode"] = ud.string(forKey: "chatBridgeMode") ?? "agentSDK"
+
         // -- UI Preferences --
         props["multi_chat_enabled"] = ud.bool(forKey: "multiChatEnabled")
         props["conversations_compact_view"] = ud.object(forKey: "conversationsCompactView") as? Bool ?? true
@@ -819,6 +827,9 @@ class AnalyticsManager {
         props["floating_bar_enabled"] = FloatingControlBarManager.shared.isEnabled
         props["floating_bar_visible"] = FloatingControlBarManager.shared.isVisible
 
+        // -- Dev Mode --
+        props["dev_mode_enabled"] = ud.bool(forKey: "devModeEnabled")
+
         return props
     }
 

desktop/Desktop/Sources/AppState.swift

@@ -870,22 +870,46 @@ class AppState: ObservableObject {
         let result = AXUIElementCopyAttributeValue(appElement, kAXFocusedWindowAttribute as CFString, &focusedWindow)
 
         // .success or .noValue (app has no windows) both mean AX is working
-        // .cannotComplete or .apiDisabled mean the permission is stuck
         switch result {
         case .success, .noValue, .notImplemented, .attributeUnsupported:
             return true
         case .apiDisabled:
+            // System-wide AX is disabled — unambiguous, no confirmation needed
             log("ACCESSIBILITY_CHECK: AXError.apiDisabled — permission stuck (tested against pid \(frontApp.processIdentifier), app: \(frontApp.localizedName ?? "unknown"))")
             return false
         case .cannotComplete:
-            log("ACCESSIBILITY_CHECK: AXError.cannotComplete — permission may be stuck (tested against pid \(frontApp.processIdentifier), app: \(frontApp.localizedName ?? "unknown"))")
-            return false
+            // cannotComplete is ambiguous: it can mean our permission is broken, OR that the
+            // frontmost app doesn't implement AX (e.g. Qt, OpenGL, Python-based apps like PyMOL).
+            // Confirm against Finder before concluding the permission is truly broken.
+            return confirmAccessibilityBrokenViaFinder(suspectApp: frontApp.localizedName ?? "unknown")
         default:
             log("ACCESSIBILITY_CHECK: AXError code \(result.rawValue) from app \(frontApp.localizedName ?? "unknown") — not permission-related, treating as OK")
             return true
         }
     }
 
+    /// Secondary AX check against Finder to disambiguate cannotComplete errors.
+    /// If Finder (a known AX-compliant app) also fails, the permission is truly broken.
+    /// If Finder succeeds, the original failure was app-specific, not a permission issue.
+    private func confirmAccessibilityBrokenViaFinder(suspectApp: String) -> Bool {
+        if let finder = NSRunningApplication.runningApplications(withBundleIdentifier: "com.apple.finder").first {
+            let finderElement = AXUIElementCreateApplication(finder.processIdentifier)
+            var finderWindow: CFTypeRef?
+            let finderResult = AXUIElementCopyAttributeValue(finderElement, kAXFocusedWindowAttribute as CFString, &finderWindow)
+            if finderResult == .cannotComplete || finderResult == .apiDisabled {
+                log("ACCESSIBILITY_CHECK: AXError.cannotComplete confirmed by Finder — permission is truly stuck (original app: \(suspectApp))")
+                return false
+            } else {
+                log("ACCESSIBILITY_CHECK: AXError.cannotComplete from \(suspectApp) but Finder OK — app-specific AX incompatibility, permission is fine")
+                return true
+            }
+        } else {
+            // Finder not running — fall back to event tap probe as tie-breaker
+            log("ACCESSIBILITY_CHECK: AXError.cannotComplete from \(suspectApp), Finder not running — using event tap probe")
+            return probeAccessibilityViaEventTap()
+        }
+    }
+
     /// Probe accessibility permission by attempting to create a CGEvent tap.
     /// Unlike AXIsProcessTrusted(), event tap creation checks the live TCC database,
     /// bypassing the per-process cache that can go stale on macOS 26 (Tahoe).
@@ -1106,6 +1130,7 @@ class AppState: ObservableObject {
             )
 
             isTranscribing = true
+            AssistantSettings.shared.transcriptionEnabled = true
             audioSource = effectiveSource
             currentTranscript = ""
             speakerSegments = []
@@ -1170,12 +1195,12 @@ class AppState: ObservableObject {
             await startBleAudioCapture()
         } else {
             // Use microphone (+ optional system audio)
-            startMicrophoneAudioCapture()
+            await startMicrophoneAudioCapture()
         }
     }
 
     /// Start microphone audio capture (original implementation)
-    private func startMicrophoneAudioCapture() {
+    private func startMicrophoneAudioCapture() async {
         guard let audioCaptureService = audioCaptureService,
               let audioMixer = audioMixer else { return }
 
@@ -1186,7 +1211,7 @@ class AppState: ObservableObject {
 
         do {
             // Start microphone capture - sends to mixer channel 0 (left/user)
-            try audioCaptureService.startCapture(
+            try await audioCaptureService.startCapture(
                 onAudioChunk: { [weak self] audioData in
                     self?.audioMixer?.setMicAudio(audioData)
                 },
@@ -1201,7 +1226,7 @@ class AppState: ObservableObject {
             if #available(macOS 14.4, *) {
                 if let systemService = systemAudioCaptureService as? SystemAudioCaptureService {
                     do {
-                        try systemService.startCapture(
+                        try await systemService.startCapture(
                             onAudioChunk: { [weak self] audioData in
                                 self?.audioMixer?.setSystemAudio(audioData)
                             },
@@ -1615,8 +1640,14 @@ class AppState: ObservableObject {
                 }
             }
         } catch {
-            // Silently ignore errors during auto-refresh — cached data stays visible
-            logError("Conversations: Auto-refresh failed", error: error)
+            // Silently ignore errors during auto-refresh — cached data stays visible.
+            // Auth errors (notSignedIn) are transient: token refresh may fail momentarily
+            // while the user is still signed in. Don't send these to Sentry.
+            if case AuthError.notSignedIn = error {
+                log("Conversations: Auto-refresh skipped (auth token temporarily unavailable)")
+            } else {
+                logError("Conversations: Auto-refresh failed", error: error)
+            }
         }
 
         // Update total count
@@ -2489,30 +2520,32 @@ class AppState: ObservableObject {
         // Create a test capture service
         let testService = SystemAudioCaptureService()
 
-        do {
-            // Try to start capture - this will fail if permission is not granted
-            try testService.startCapture { _ in
-                // We don't need the audio data, just testing if it works
-            }
+        Task {
+            do {
+                // Try to start capture - this will fail if permission is not granted
+                try await testService.startCapture { _ in
+                    // We don't need the audio data, just testing if it works
+                }
 
-            // If we get here, capture started successfully
-            log("System audio: Test capture started successfully")
+                // If we get here, capture started successfully
+                log("System audio: Test capture started successfully")
 
-            // Stop the test capture
-            testService.stopCapture()
-            log("System audio: Test capture stopped")
+                // Stop the test capture
+                testService.stopCapture()
+                log("System audio: Test capture stopped")
 
-            // Mark permission as granted
-            hasSystemAudioPermission = true
-            log("System audio: Permission verified")
+                // Mark permission as granted
+                hasSystemAudioPermission = true
+                log("System audio: Permission verified")
 
-        } catch {
-            logError("System audio: Test capture failed", error: error)
-            hasSystemAudioPermission = false
+            } catch {
+                logError("System audio: Test capture failed", error: error)
+                hasSystemAudioPermission = false
 
-            // Open System Settings to Screen Recording section
-            if let url = URL(string: "x-apple.systempreferences:com.apple.preference.security?Privacy_ScreenCapture") {
-                NSWorkspace.shared.open(url)
+                // Open System Settings to Screen Recording section
+                if let url = URL(string: "x-apple.systempreferences:com.apple.preference.security?Privacy_ScreenCapture") {
+                    NSWorkspace.shared.open(url)
+                }
             }
         }
     }

desktop/Desktop/Sources/Audio/AudioSourceManager.swift

@@ -227,7 +227,7 @@ final class AudioSourceManager: ObservableObject {
         }
 
         // Start microphone capture
-        try audioCaptureService?.startCapture(
+        try await audioCaptureService?.startCapture(
             onAudioChunk: { [weak self] audioData in
                 self?.audioMixer?.setMicAudio(audioData)
             },
@@ -241,7 +241,7 @@ final class AudioSourceManager: ObservableObject {
 
         // Start system audio capture if available
         if #available(macOS 14.4, *), let systemCapture = systemAudioCaptureService as? SystemAudioCaptureService {
-            try systemCapture.startCapture(
+            try await systemCapture.startCapture(
                 onAudioChunk: { [weak self] audioData in
                     self?.audioMixer?.setSystemAudio(audioData)
                 },

desktop/Desktop/Sources/AudioCaptureService.swift

@@ -6,7 +6,7 @@ import CoreAudio
 /// Uses CoreAudio IOProc directly on the default input device to avoid
 /// AVAudioEngine's implicit aggregate device creation, which degrades
 /// system audio output quality (especially Bluetooth A2DP → SCO switch).
-class AudioCaptureService {
+class AudioCaptureService: @unchecked Sendable {
 
     // MARK: - Types
 
@@ -105,7 +105,7 @@ class AudioCaptureService {
     /// - Parameters:
     ///   - onAudioChunk: Callback receiving 16-bit PCM audio data chunks at 16kHz
     ///   - onAudioLevel: Optional callback receiving normalized audio level (0.0 - 1.0)
-    func startCapture(onAudioChunk: @escaping AudioChunkHandler, onAudioLevel: AudioLevelHandler? = nil) throws {
+    func startCapture(onAudioChunk: @escaping AudioChunkHandler, onAudioLevel: AudioLevelHandler? = nil) async throws {
         guard !isCapturing else {
             log("AudioCapture: Already capturing")
             return
@@ -114,6 +114,28 @@ class AudioCaptureService {
         self.onAudioChunk = onAudioChunk
         self.onAudioLevel = onAudioLevel
 
+        // All CoreAudio HAL calls (AudioObjectGetPropertyData, AudioDeviceStart, etc.) are
+        // synchronous IPC to coreaudiod via mach_msg. After wake from sleep the daemon can
+        // take seconds to respond, blocking the caller. Dispatch the entire setup to audioQueue,
+        // mirroring the pattern already used in stopCapture() and handleConfigurationChange().
+        try await withCheckedThrowingContinuation { (continuation: CheckedContinuation<Void, Error>) in
+            audioQueue.async { [weak self] in
+                guard let self else {
+                    continuation.resume()
+                    return
+                }
+                do {
+                    try self.startCaptureOnQueue()
+                    continuation.resume()
+                } catch {
+                    continuation.resume(throwing: error)
+                }
+            }
+        }
+    }
+
+    /// Performs all blocking CoreAudio HAL setup. Must be called on audioQueue, not the main thread.
+    private func startCaptureOnQueue() throws {
         // 1. Get default input device
         var inputDeviceID: AudioDeviceID = kAudioObjectUnknown
         var size = UInt32(MemoryLayout<AudioDeviceID>.size)