feat: add offline meeting transcription milestone 1

Info.plist

@@ -24,6 +24,10 @@
 	<string>FluidVoice needs Accessibility permission to listen for global hotkeys and type text into other apps when you stop recording.</string>
 	<key>NSMicrophoneUsageDescription</key>
 	<string>FluidVoice needs microphone access to record your voice for speech-to-text transcription. Your audio is processed locally and never sent to external servers.</string>
+	<key>NSAudioCaptureUsageDescription</key>
+	<string>FluidVoice needs system audio access to record the meeting application you explicitly select. Meeting audio stays on this Mac.</string>
+	<key>NSScreenCaptureUsageDescription</key>
+	<string>FluidVoice needs Screen &amp; System Audio access to record audio from the meeting application you explicitly select.</string>
 	<key>LSUIElement</key>
 	<false/>
 	<key>NSAppleEventsUsageDescription</key>

MEETING_TRANSCRIPTION_PRD.md

@@ -25,7 +25,7 @@ Checkboxes should be checked only when the requirement and its acceptance criter
 
 ### Document map
 
-- [Agreed product requirements](#3-agreed-product-requirements--implementation-status)
+- [Confirmed decisions](#3-confirmed-product-decisions)
 - [Goals and non-goals](#4-goals-and-non-goals)
 - [High-level milestones](#5-high-level-milestones)
 - [User experience and canvas states](#7-user-experience-and-canvas-states)
@@ -128,6 +128,15 @@ The direction below is agreed. Its checkboxes remain open until the correspondin
 
 ## 5. High-level milestones
 
+### Milestone delivery policy
+
+Each milestone is implemented on a stacked branch targeting `1.6.8/meeting`; only the integration branch targets `main`. A milestone is not complete when code merely compiles.
+
+- [ ] **[MUST · FOUNDATION] DELIVERY-001** Define the milestone's deterministic tests, real-system proof, performance/resource checks, and regression boundaries before implementation is called complete.
+- [ ] **[MUST · FOUNDATION] DELIVERY-002** Assign independent architecture/reliability and UX/accessibility reviewers who did not own the implementation.
+- [ ] **[MUST · FOUNDATION] DELIVERY-003** Resolve every blocking finding and record deferred non-blockers against a named later milestone.
+- [ ] **[MUST · FOUNDATION] DELIVERY-004** Merge the reviewed milestone PR into `1.6.8/meeting`; open one final integration PR from `1.6.8/meeting` to `main` after the intended milestone set is complete.
+
 ### Milestone 0 — Decisions and durable foundations
 
 - [ ] **[MUST · FOUNDATION] M0-EXIT-001** Resolve FOUNDATION and V1 open decisions in Section 22; M3/M4 decisions remain gated by their own milestones.
@@ -146,6 +155,18 @@ Goal: prove the complete path before polishing secondary automation.
 - [ ] **[MUST · M1] M1-EXIT-005** The menu-bar menu shows **Stop Meeting Recording** first while capture is active.
 - [ ] **[MUST · M1] M1-EXIT-006** Window closure and sidebar navigation do not stop the session.
 
+#### Milestone 1 test and review gate
+
+- [ ] **[MUST · M1] M1-TEST-001** Deterministic tests cover legal state transitions, concurrent/repeated Start and Stop, stale-generation callbacks, and exactly one finalization/processing job.
+- [ ] **[MUST · M1] M1-TEST-002** A real online-call capture proves separate non-empty application/system and microphone files with timestamps inside one session manifest.
+- [ ] **[MUST · M1] M1-TEST-003** A real in-room capture proves microphone-only recording does not request Screen/System Audio permission solely for that mode.
+- [ ] **[MUST · M1] M1-TEST-004** Stop proves the complete offline English path: finalized audio → transcription/diarization → timestamped canvas result.
+- [ ] **[MUST · M1] M1-TEST-005** Apple Silicon fixtures cover remote speaker separation, clean confirmed personal-mic **You**, and ambiguous/leaked microphone speech remaining unknown.
+- [ ] **[MUST · M1] M1-TEST-006** UI proof covers menu item order and Stop action, menu-bar recording mark, sidebar navigation, main-window close/reopen, and the four canvas states.
+- [ ] **[MUST · M1] M1-REVIEW-001** An independent audio/architecture reviewer approves capture ownership, track separation, timestamp handling, provider serialization, and no dictation first-PCM regression.
+- [ ] **[MUST · M1] M1-REVIEW-002** An independent UX/accessibility reviewer approves sidebar placement, state clarity, menu-bar behavior, keyboard/VoiceOver labels, and low-resource rendering.
+- [ ] **[MUST · M1] M1-REVIEW-003** Every blocking review finding is fixed or explicitly recorded as a later-milestone item before the M1 stacked PR merges.
+
 ### Milestone 2 — Trustworthy offline V1
 
 Goal: make the working slice safe to ship.
@@ -719,8 +740,8 @@ flowchart TD
 - [ ] **[MUST · V1] TEST-REPO-002** Validate implementation builds with `sh build_with_FI_incremental.sh` and inspect the installed `/Applications/FluidVoice.app`.
 - [ ] **[MUST · V1] TEST-REPO-003** Do not treat Debug/XCTest-host runs as release proof; use focused logic checks plus installed-app, real-meeting evidence.
 - [ ] **[MUST · V1] TEST-REPO-004** Verify the minimum supported macOS path and low-resource Apple Silicon behavior; describe Intel as reduced capability until separately proven.
-- [ ] **[MUST · V1] TEST-REPO-005** Update the requested release-notes version locally, credit contributors as required, and never commit ignored release notes.
 - [ ] **[MUST · V1] TEST-REPO-006** Run real Intel capture plus plain English transcription before claiming Intel meeting support; universal build slices alone are insufficient.
+- [ ] **[MUST · V1] TEST-REPO-005** Update the requested release-notes version locally, credit contributors as required, and never commit ignored release notes.
 
 ### 21.7 Measurable budgets to set before implementation
 

Sources/Fluid/AppDelegate.swift

@@ -89,7 +89,9 @@ class AppDelegate: NSObject, NSApplicationDelegate, UNUserNotificationCenterDele
             didFinishShutdown = true
         }
 
-        let deadline = Date().addingTimeInterval(8)
+        // Meeting capture can spend up to three seconds stopping its runtime and
+        // four seconds finalizing audio before the durable session save.
+        let deadline = Date().addingTimeInterval(12)
         while !didFinishShutdown, Date() < deadline {
             RunLoop.current.run(mode: .default, before: Date().addingTimeInterval(0.01))
         }

Sources/Fluid/ContentView.swift

@@ -78,7 +78,8 @@ enum SidebarItem: Hashable {
     case voiceEngine
     case aiEnhancements
     case preferences
-    case meetingTools
+    case fileTranscription
+    case meetingTranscription
     case customDictionary
     case stats
     case history
@@ -591,6 +592,9 @@ struct ContentView: View {
         DispatchQueue.main.asyncAfter(deadline: .now() + 1.5) {
             DebugLogger.shared.info("🚦 Startup delay complete, signaling UI ready...", source: "ContentView")
             self.appServices.signalUIReady()
+            self.menuBarManager.configure(
+                meetingCoordinator: self.appServices.meetingSessionCoordinator
+            )
 
             Task { @MainActor in
                 await AudioStartupGate.shared.scheduleOpenAfterInitialUISettled()
@@ -958,6 +962,8 @@ struct ContentView: View {
         switch destination {
         case .customDictionary:
             self.selectedSidebarItem = .customDictionary
+        case .meetingTranscription:
+            self.selectedSidebarItem = .meetingTranscription
         case .preferences:
             self.selectedSidebarItem = .preferences
         }
@@ -1166,7 +1172,12 @@ struct ContentView: View {
 
             Section {
                 self.sidebarNavigationLink(.commandMode, title: "Command Mode", systemImage: "terminal.fill")
-                self.sidebarNavigationLink(.meetingTools, title: "File Transcription", systemImage: "doc.text.fill")
+                self.sidebarNavigationLink(.fileTranscription, title: "File Transcription", systemImage: "doc.text.fill")
+                self.sidebarNavigationLink(
+                    .meetingTranscription,
+                    title: "Meeting Transcription",
+                    systemImage: "person.2.wave.2.fill"
+                )
             } header: {
                 self.sidebarSectionHeader("Use")
             }
@@ -1264,8 +1275,10 @@ struct ContentView: View {
             ))
         case .preferences:
             return AnyView(self.preferencesView)
-        case .meetingTools:
-            return AnyView(self.meetingToolsView)
+        case .fileTranscription:
+            return AnyView(self.fileTranscriptionView)
+        case .meetingTranscription:
+            return AnyView(self.meetingTranscriptionView)
         case .customDictionary:
             return AnyView(CustomDictionaryView())
         case .stats:
@@ -1500,10 +1513,21 @@ struct ContentView: View {
         })
     }
 
-    // MARK: - Meeting Transcription (Coming Soon)
+    // MARK: - File and Meeting Transcription
+
+    private var fileTranscriptionView: some View {
+        FileTranscriptionView(
+            asrService: self.asr,
+            transcriptionService: self.appServices.fileTranscriptionService
+        )
+    }
 
-    private var meetingToolsView: some View {
-        MeetingTranscriptionView(asrService: self.asr)
+    private var meetingTranscriptionView: some View {
+        MeetingTranscriptionView(
+            coordinator: self.appServices.meetingSessionCoordinator,
+            asrService: self.asr,
+            onOpenVoiceEngine: { self.selectedSidebarItem = .voiceEngine }
+        )
     }
 
     // MARK: - Stats View
@@ -3068,6 +3092,7 @@ struct ContentView: View {
 
     /// Capture app context at start to avoid mismatches if the user switches apps mid-session
     private func startRecording() {
+        guard !self.presentExclusiveActivityBlockIfNeeded() else { return }
         let model = SettingsStore.shared.selectedSpeechModel
         DebugLogger.shared.info(
             "ContentView: startRecording() for model=\(model.displayName), supportsStreaming=\(model.supportsStreaming)",
@@ -3077,7 +3102,6 @@ struct ContentView: View {
             DebugLogger.shared.debug("ContentView: start ignored because capture is already active", source: "ContentView")
             return
         }
-
         self.advanceOverlayLifecycle()
         self.setActiveRecordingMode(.dictate)
         let shouldShowDictationOverlay = !self.isRecordingForCommand
@@ -3128,6 +3152,20 @@ struct ContentView: View {
         }
     }
 
+    private func presentExclusiveActivityBlockIfNeeded() -> Bool {
+        guard let activity = self.asr.activeExclusiveActivity else { return false }
+        let message = "Wait for the active \(activity.displayName) to finish."
+        self.asr.errorTitle = "Dictation Unavailable"
+        self.asr.errorMessage = message
+        self.asr.showError = true
+        AccessibilityNotification.Announcement("Dictation unavailable. \(message)").post()
+        DebugLogger.shared.info(
+            "ContentView: dictation start blocked by \(activity.rawValue)",
+            source: "ContentView"
+        )
+        return true
+    }
+
     private func prewarmPrivateAIDictationIfNeeded(for slot: SettingsStore.DictationShortcutSlot) {
         let appBundleID = self.recordingAppInfo?.bundleId
         guard PrivateAIProviderPromptFormat.isAvailable(settings: SettingsStore.shared),
@@ -3337,6 +3375,7 @@ struct ContentView: View {
                 self.beginDictationRecording(for: selection, mode: .promptMode)
             },
             commandModeCallback: {
+                guard !self.presentExclusiveActivityBlockIfNeeded() else { return }
                 DebugLogger.shared.info("Command mode triggered", source: "ContentView")
                 self.captureRecordingContext()
 
@@ -3368,6 +3407,7 @@ struct ContentView: View {
                 }
             },
             rewriteModeCallback: {
+                guard !self.presentExclusiveActivityBlockIfNeeded() else { return }
                 guard !self.showPrivateAIEditModeUnavailableIfNeeded() else { return }
 
                 self.captureRecordingContext()
@@ -3721,6 +3761,7 @@ extension ContentView {
     }
 
     private func beginDictationRecording(for slot: SettingsStore.DictationShortcutSlot, mode: ActiveRecordingMode) {
+        guard !self.presentExclusiveActivityBlockIfNeeded() else { return }
         DebugLogger.shared.debug("Begin dictation recording for slot \(slot.rawValue)", source: "ContentView")
         self.appBench("begin_recording slot=\(slot.rawValue) mode=\(mode.rawValue)")
         if self.isOnboardingVoicePlaygroundStepActive {

Sources/Fluid/Persistence/SettingsStore.swift

@@ -1813,6 +1813,23 @@ final class SettingsStore: ObservableObject {
         set { self.defaults.set(newValue, forKey: Keys.preferredInputDeviceUID) }
     }
 
+    var meetingRecordingDefaults: MeetingRecordingDefaults {
+        get {
+            guard let data = self.defaults.data(forKey: Keys.meetingRecordingDefaults),
+                  let saved = try? JSONDecoder().decode(MeetingRecordingDefaults.self, from: data)
+            else {
+                return .unconfigured
+            }
+            return saved
+        }
+        set {
+            guard let data = try? JSONEncoder().encode(newValue) else {
+                return
+            }
+            self.defaults.set(data, forKey: Keys.meetingRecordingDefaults)
+        }
+    }
+
     var preferredOutputDeviceUID: String? {
         get { self.defaults.string(forKey: Keys.preferredOutputDeviceUID) }
         set { self.defaults.set(newValue, forKey: Keys.preferredOutputDeviceUID) }
@@ -4885,6 +4902,7 @@ private extension SettingsStore {
         static let primaryDictationShortcutsKey = "PrimaryDictationShortcuts"
         static let preferredInputDeviceUID = "PreferredInputDeviceUID"
         static let preferredOutputDeviceUID = "PreferredOutputDeviceUID"
+        static let meetingRecordingDefaults = "MeetingRecordingDefaults"
         static let microphoneSelectionMode = "MicrophoneSelectionMode"
         static let appMicSelectionMigrationVersion = "AppOnlyMicrophoneSelectionMigrationVersion"
         static let visualizerNoiseThreshold = "VisualizerNoiseThreshold"