refactor(desktop): realtime hub as default voice path + revert pill UI

desktop/macos/CHANGELOG.json

@@ -1,10 +1,10 @@
 {
   "unreleased": [
     "Faster, cheaper assistant responses via Anthropic prompt caching of the system+tools prefix and conversation history",
-    "Added an experimental Realtime Voice Hub (Settings \u2192 Floating Bar): the realtime model handles your whole voice turn \u2014 listening, deciding, and speaking \u2014 for noticeably faster replies",
-    "Redesigned the floating bar voice indicator with smooth, distinct idle, listening, thinking, and speaking states so you always know whether the assistant is working or done",
+    "Faster voice replies (experimental): the realtime model now handles your whole voice turn \u2014 listening, deciding, and speaking \u2014 instead of the slower transcribe\u2192route\u2192answer pipeline",
     "Voice (push-to-talk) conversations now appear in your chat history",
-    "Fixed older chat messages failing to load in long chats"
+    "Fixed older chat messages failing to load in long chats",
+    "Replaced the push-to-talk red dot in the floating bar with a realtime audio equalizer"
   ],
   "releases": [
     {

desktop/macos/Desktop/Sources/FloatingControlBar/AIResponseView.swift

@@ -349,11 +349,8 @@ struct AIResponseView: View {
 
     private var voiceFollowUpView: some View {
         HStack(spacing: 8) {
-            Circle()
-                .fill(Color.red)
-                .frame(width: 10, height: 10)
-                .scaleEffect(1.2)
-                .animation(.easeInOut(duration: 0.6).repeatForever(autoreverses: true), value: isVoiceFollowUp)
+            // Playful realtime mic waveform (replaces the old pulsing red dot)
+            VoiceWaveformBars(isActive: isVoiceFollowUp)
 
             Image(systemName: "mic.fill")
                 .scaledFont(size: 14, weight: .semibold)
@@ -375,7 +372,7 @@ struct AIResponseView: View {
         }
         .padding(.horizontal, 10)
         .padding(.vertical, 8)
-        .background(Color.red.opacity(0.15))
+        .background(OmiColors.purplePrimary.opacity(0.12))
         .cornerRadius(8)
     }
 

desktop/macos/Desktop/Sources/FloatingControlBar/AboutUserCard.swift

@@ -0,0 +1,51 @@
+import Foundation
+
+/// Builds the compact, local-only `<about_user>` block injected into the hub's
+/// system instruction at warm time. Identity + rough situation only; exact/current
+/// lists stay behind the read tools (the card hedges this). No network calls.
+enum AboutUserCard {
+  /// Pure formatter — kept separate from `build()` so it is unit-testable.
+  static func render(name: String, facts: [String], overdue: Int, dueToday: Int) -> String {
+    var lines: [String] = ["<about_user>"]
+    if !name.isEmpty { lines.append("Name: \(name)") }
+    lines.append("What Omi knows about them:")
+    if facts.isEmpty {
+      lines.append("- Nothing saved yet.")
+    } else {
+      lines.append(contentsOf: facts.map { "- \($0)" })
+    }
+    if overdue == 0 && dueToday == 0 {
+      lines.append("Right now: nothing overdue or due today.")
+    } else {
+      lines.append("Right now: \(overdue) overdue, \(dueToday) due today.")
+    }
+    lines.append(
+      "(This is a quick snapshot — for the exact or current list, call get_tasks / get_action_items.)")
+    lines.append("</about_user>")
+    return lines.joined(separator: "\n")
+  }
+
+  /// Gathers local data (auth name, top memories, task counts) and renders the card.
+  /// Best-effort: any failure degrades to a smaller card, never throws.
+  @MainActor
+  static func build() async -> String {
+    let auth = AuthService.shared
+    let rawName = auth.givenName.isEmpty ? auth.displayName : auth.givenName
+    let name = rawName.trimmingCharacters(in: .whitespacesAndNewlines)
+
+    var facts: [String] = []
+    if let memories = try? await MemoryStorage.shared.getLocalMemories(limit: 8) {
+      facts = memories.prefix(8).compactMap { mem in
+        let t = mem.content.trimmingCharacters(in: .whitespacesAndNewlines)
+        guard !t.isEmpty else { return nil }
+        return t.count > 120 ? String(t.prefix(117)) + "…" : t
+      }
+    }
+
+    await TasksStore.shared.loadDashboardTasks()
+    let overdue = TasksStore.shared.overdueTasks.count
+    let dueToday = TasksStore.shared.todaysTasks.count
+
+    return render(name: name, facts: facts, overdue: overdue, dueToday: dueToday)
+  }
+}

desktop/macos/Desktop/Sources/FloatingControlBar/FloatingControlBarState.swift

@@ -51,21 +51,6 @@ struct FloatingBarNotification: Identifiable, Equatable {
     }
 }
 
-/// The high-level voice activity the floating bar is reflecting right now. Derived
-/// from the lower-level PTT/hub flags so the status indicator has a single, ordered
-/// source of truth (each state has exactly one visual treatment).
-enum VoiceActivity: Equatable {
-    /// Nothing happening — the bar rests as a calm, barely-breathing sliver.
-    case idle
-    /// User is holding push-to-talk; we're capturing their voice (red, "you").
-    case listening
-    /// Turn committed, waiting on the model's reply — the model may answer late,
-    /// so this MUST read as "working, wait" rather than "done" (cool autonomous swirl).
-    case thinking
-    /// The model is speaking its reply (warm, audio-reactive waveform — "it").
-    case speaking
-}
-
 /// Observable object holding the state for the floating control bar.
 @MainActor
 class FloatingControlBarState: NSObject, ObservableObject {
@@ -108,35 +93,6 @@ class FloatingControlBarState: NSObject, ObservableObject {
     @Published var isVoiceListening: Bool = false
     @Published var isVoiceLocked: Bool = false
     @Published var voiceTranscript: String = ""
-    /// True after a voice turn is committed and we're waiting on the model's reply
-    /// (vs. still recording) — drives the "Thinking…/Responding…" indicator so the user
-    /// knows to wait rather than re-pressing (which would interrupt a slow reply).
-    @Published var isVoiceThinking: Bool = false
-    /// True while the model is actually speaking its reply (native audio playing or the
-    /// AVSpeech fallback talking). Distinct from `isVoiceThinking` so the indicator can
-    /// show a clearly different "it's talking" treatment vs. "it's working".
-    @Published var isVoiceSpeaking: Bool = false
-    /// Smoothed 0…1 output amplitude of the model's spoken reply, sampled from the
-    /// playback engine. Drives the speaking waveform so it reacts to the actual voice
-    /// (premium feel) rather than animating blindly. 0 when not speaking.
-    @Published var voiceLevel: CGFloat = 0
-
-    /// Single ordered source of truth for the status indicator. Listening wins (the user
-    /// is actively talking), then speaking, then thinking, else idle — by construction the
-    /// hub sets these mutually exclusively, the ordering just makes barge-in race-safe.
-    var voiceActivity: VoiceActivity {
-        if isVoiceListening { return .listening }
-        if isVoiceSpeaking { return .speaking }
-        if isVoiceThinking { return .thinking }
-        return .idle
-    }
-
-    /// Whether any voice turn is in flight — keeps the bar expanded across the whole
-    /// listening → thinking → speaking arc so the indicator stays visible (one expand,
-    /// one collapse per turn — no resize churn mid-turn).
-    var isVoiceActive: Bool {
-        isVoiceListening || isVoiceThinking || isVoiceSpeaking
-    }
 
     // Voice follow-up state (PTT while AI conversation is active)
     @Published var isVoiceFollowUp: Bool = false
@@ -180,8 +136,6 @@ class FloatingControlBarState: NSObject, ObservableObject {
         isVoiceFollowUp = false
         voiceFollowUpTranscript = ""
         currentQueryFromVoice = false
-        isVoiceSpeaking = false
-        voiceLevel = 0
         lastConversationActivityAt = nil
     }
 }

desktop/macos/Desktop/Sources/FloatingControlBar/FloatingControlBarView.swift

@@ -33,11 +33,9 @@ struct FloatingControlBarView: View {
         .animation(.spring(response: 0.35, dampingFraction: 0.82), value: state.currentNotification?.id)
     }
 
-    /// Whether the bar chrome should stretch to fill the window width. Stays full-width
-    /// for the whole voice turn (listening → thinking → speaking) so the status indicator
-    /// has room and the bar resizes exactly once per turn.
+    /// Whether the bar chrome should stretch to fill the window width
     private var barNeedsFullWidth: Bool {
-        isHovering || state.showingAIConversation || state.isVoiceActive
+        isHovering || state.showingAIConversation || state.isVoiceListening
     }
 
     private var barChrome: some View {
@@ -85,7 +83,7 @@ struct FloatingControlBarView: View {
             }
         }
         .overlay(alignment: .topTrailing) {
-            if isHovering && !state.isVoiceActive {
+            if isHovering && !state.isVoiceListening {
                 Button {
                     openFloatingBarSettings()
                 } label: {
@@ -281,8 +279,8 @@ struct FloatingControlBarView: View {
 
     private var controlBarView: some View {
         Group {
-            if state.isVoiceActive && !state.isVoiceFollowUp {
-                voiceActiveView
+            if state.isVoiceListening && !state.isVoiceFollowUp {
+                voiceListeningView
                     .padding(.horizontal, 6)
                     .padding(.vertical, 3)
                     .frame(height: 50)
@@ -308,11 +306,11 @@ struct FloatingControlBarView: View {
         }
     }
 
-    /// Minimal resting indicator shown when not hovering and no voice turn is active —
-    /// a calm, slowly breathing sliver. (Active turns render `voiceActiveView` instead.)
+    /// Minimal thin bar shown when not hovering
     private var compactCircleView: some View {
-        VoiceActivityIndicator(activity: state.voiceActivity, level: state.voiceLevel)
-            .frame(width: 28, height: 14)
+        RoundedRectangle(cornerRadius: 3)
+            .fill(Color.white.opacity(0.5))
+            .frame(width: 28, height: 6)
     }
 
     private func compactToggle(_ title: String, isOn: Binding<Bool>) -> some View {
@@ -360,15 +358,16 @@ struct FloatingControlBarView: View {
         }
     }
 
-    /// Unified expanded voice view for the whole turn. The status indicator carries the
-    /// state (listening / thinking / speaking) visually; the text is just the helpful
-    /// detail (transcript, "Release to send", "Thinking…"). One element, no jarring swaps.
-    private var voiceActiveView: some View {
+    private var voiceListeningView: some View {
         HStack(spacing: 8) {
-            VoiceActivityIndicator(activity: state.voiceActivity, level: state.voiceLevel)
-                .frame(width: 34, height: 18)
+            // Playful realtime mic waveform (replaces the old pulsing red dot)
+            VoiceWaveformBars(isActive: state.isVoiceListening)
 
-            if state.isVoiceLocked && state.isVoiceListening {
+            Image(systemName: "mic.fill")
+                .scaledFont(size: 14, weight: .semibold)
+                .foregroundColor(.white)
+
+            if state.isVoiceLocked {
                 Text("LOCKED")
                     .scaledFont(size: 10, weight: .bold)
                     .foregroundColor(.orange)
@@ -378,31 +377,21 @@ struct FloatingControlBarView: View {
                     .cornerRadius(4)
             }
 
-            // Dim only the "Release to send" hint; live transcript / status reads brighter.
-            let isHint = state.voiceActivity == .listening && state.voiceTranscript.isEmpty
-            Text(voiceStatusText)
-                .scaledFont(size: 13)
-                .foregroundColor(.white.opacity(isHint ? 0.5 : 0.85))
-                .lineLimit(1)
-                .truncationMode(.head)
-        }
-    }
-
-    /// The detail text beside the indicator for the current voice state. The indicator
-    /// itself carries the state visually; this is just the helpful detail.
-    private var voiceStatusText: String {
-        switch state.voiceActivity {
-        case .listening:
-            if !state.voiceTranscript.isEmpty { return state.voiceTranscript }
-            return state.isVoiceLocked
-                ? "Tap \(shortcutSettings.pttShortcut.displayLabel) to send"
-                : "Release \(shortcutSettings.pttShortcut.displayLabel) to send"
-        case .thinking:
-            return "Thinking…"
-        case .speaking:
-            return "Speaking…"
-        case .idle:
-            return ""
+            if !state.voiceTranscript.isEmpty {
+                Text(state.voiceTranscript)
+                    .scaledFont(size: 13)
+                    .foregroundColor(.white.opacity(0.8))
+                    .lineLimit(1)
+                    .truncationMode(.head)
+            } else {
+                Text(
+                    state.isVoiceLocked
+                        ? "Tap \(shortcutSettings.pttShortcut.displayLabel) to send"
+                        : "Release \(shortcutSettings.pttShortcut.displayLabel) to send"
+                )
+                    .scaledFont(size: 13)
+                    .foregroundColor(.white.opacity(0.5))
+            }
         }
     }