myVoxtral/docs/superpowers/plans/2026-04-07-myvoxtral.md

MyVoxtral Implementation Plan

For agentic workers: REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (- [ ]) syntax for tracking.

Goal: Build a macOS menu bar app that streams microphone audio to Mistral Voxtral Realtime API and outputs transcribed text to a floating window or the active cursor.

Architecture: SwiftUI menu bar app with four layers — AudioCapture (AVAudioEngine), VoxtralWebSocketClient (URLSessionWebSocketTask), TranscriptionManager (ObservableObject orchestrator), and UI (MenuBarExtra + floating panel). No third-party dependencies.

Tech Stack: Swift, SwiftUI, AVFoundation, WebSocket (URLSession), CGEvent, macOS 14+

---

File Structure

MyVoxtral/
├── MyVoxtralApp.swift              # App entry point, MenuBarExtra, app lifecycle
├── Models/
│   ├── TranscriptionManager.swift  # State machine orchestrator, owns audio + WS client
│   └── AppSettings.swift           # @AppStorage wrapper, settings model
├── Audio/
│   └── AudioCapture.swift          # AVAudioEngine mic tap, PCM conversion, async chunk stream
├── Network/
│   ├── VoxtralWebSocketClient.swift # WebSocket connect/send/receive, session management
│   └── VoxtralMessages.swift       # Codable structs for all WS JSON messages
├── Views/
│   ├── TranscriptionWindow.swift   # NSPanel-based floating text window
│   ├── SettingsView.swift          # API key, shortcut, mode, latency
│   └── MenuBarView.swift           # Menu bar dropdown content
├── Utilities/
│   ├── CursorInjector.swift        # CGEvent keystroke simulation
│   ├── GlobalShortcut.swift        # NSEvent global monitor, key combo storage
│   └── TranscriptionLogger.swift   # Append sessions to log file
└── Info.plist                       # Mic + accessibility usage descriptions

---

Task 1: Xcode Project Scaffold

Files:

• Create: MyVoxtral.xcodeproj (via swift package init or Xcode project)

• Create: MyVoxtral/MyVoxtralApp.swift

• Create: MyVoxtral/Info.plist

• Step 1: Create the Swift package / Xcode project

Create a new macOS app project. Use SwiftUI lifecycle.

mkdir -p MyVoxtral/MyVoxtral

Create MyVoxtral/MyVoxtral/MyVoxtralApp.swift:

import SwiftUI

@main
struct MyVoxtralApp: App {
    var body: some Scene {
        MenuBarExtra("MyVoxtral", systemImage: "mic.fill") {
            Text("MyVoxtral")
            Divider()
            Button("Quit") {
                NSApplication.shared.terminate(nil)
            }
        }
    }
}

• Step 2: Create Info.plist with privacy descriptions

Create MyVoxtral/MyVoxtral/Info.plist:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>NSMicrophoneUsageDescription</key>
    <string>MyVoxtral needs microphone access to transcribe your speech in real time.</string>
    <key>LSUIElement</key>
    <true/>
</dict>
</plist>

LSUIElement = true makes it a menu-bar-only app (no Dock icon).

• Step 3: Create the Xcode project file

cd MyVoxtral
cat > Package.swift << 'SWIFT'
// swift-tools-version: 5.10
import PackageDescription

let package = Package(
    name: "MyVoxtral",
    platforms: [.macOS(.v14)],
    targets: [
        .executableTarget(
            name: "MyVoxtral",
            path: "MyVoxtral"
        )
    ]
)
SWIFT

• Step 4: Build and verify the empty shell runs

swift build

Expected: Build succeeds. Running shows a mic icon in the menu bar with "MyVoxtral" and "Quit".

• Step 5: Commit

git init
git add -A
git commit -m "chore: scaffold MyVoxtral macOS menu bar app"

---

Task 2: WebSocket Message Types

Files:

• Create: MyVoxtral/MyVoxtral/Network/VoxtralMessages.swift

• Step 1: Define all outbound message types

Create MyVoxtral/MyVoxtral/Network/VoxtralMessages.swift:

import Foundation

// MARK: - Outbound Messages (Client → Server)

struct AudioAppendMessage: Encodable {
    let type = "input_audio.append"
    let audio: String // base64-encoded PCM
}

struct AudioFlushMessage: Encodable {
    let type = "input_audio.flush"
}

struct AudioEndMessage: Encodable {
    let type = "input_audio.end"
}

struct SessionUpdateMessage: Encodable {
    let type = "session.update"
    let session: SessionConfig
}

struct SessionConfig: Encodable {
    let audioFormat: AudioFormatConfig
    let targetStreamingDelayMs: Int

    enum CodingKeys: String, CodingKey {
        case audioFormat = "audio_format"
        case targetStreamingDelayMs = "target_streaming_delay_ms"
    }
}

struct AudioFormatConfig: Encodable {
    let encoding = "pcm_s16le"
    let sampleRate = 16000

    enum CodingKeys: String, CodingKey {
        case encoding
        case sampleRate = "sample_rate"
    }
}

// MARK: - Inbound Messages (Server → Client)

enum VoxtralEvent {
    case sessionCreated
    case textDelta(String)
    case segment(text: String, start: Double, end: Double)
    case language(String)
    case done(text: String)
    case error(String)
    case unknown(String)
}

struct IncomingEvent: Decodable {
    let type: String
}

struct TextDeltaEvent: Decodable {
    let text: String
}

struct LanguageEvent: Decodable {
    let audioLanguage: String

    enum CodingKeys: String, CodingKey {
        case audioLanguage = "audio_language"
    }
}

struct SegmentEvent: Decodable {
    let text: String
    let start: Double
    let end: Double
}

struct DoneEvent: Decodable {
    let text: String
}

struct ErrorEvent: Decodable {
    let error: ErrorDetail?
}

struct ErrorDetail: Decodable {
    let message: ErrorMessage?
}

struct ErrorMessage: Decodable {
    let detail: String?
}

// MARK: - Event Parsing

func parseVoxtralEvent(from data: Data) -> VoxtralEvent {
    guard let envelope = try? JSONDecoder().decode(IncomingEvent.self, from: data) else {
        return .unknown(String(data: data, encoding: .utf8) ?? "")
    }

    switch envelope.type {
    case "session.created":
        return .sessionCreated
    case "transcription.text.delta":
        guard let e = try? JSONDecoder().decode(TextDeltaEvent.self, from: data) else { return .unknown("") }
        return .textDelta(e.text)
    case "transcription.segment":
        guard let e = try? JSONDecoder().decode(SegmentEvent.self, from: data) else { return .unknown("") }
        return .segment(text: e.text, start: e.start, end: e.end)
    case "transcription.language":
        guard let e = try? JSONDecoder().decode(LanguageEvent.self, from: data) else { return .unknown("") }
        return .language(e.audioLanguage)
    case "transcription.done":
        guard let e = try? JSONDecoder().decode(DoneEvent.self, from: data) else { return .unknown("") }
        return .done(text: e.text)
    case "error":
        if let e = try? JSONDecoder().decode(ErrorEvent.self, from: data) {
            return .error(e.error?.message?.detail ?? "Unknown error")
        }
        return .error("Unknown error")
    default:
        return .unknown(envelope.type)
    }
}

• Step 2: Build to verify it compiles

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Network/VoxtralMessages.swift
git commit -m "feat: add Voxtral WebSocket message types and parser"

---

Task 3: WebSocket Client

Files:

• Create: MyVoxtral/MyVoxtral/Network/VoxtralWebSocketClient.swift

• Step 1: Implement the WebSocket client

Create MyVoxtral/MyVoxtral/Network/VoxtralWebSocketClient.swift:

import Foundation

@MainActor
final class VoxtralWebSocketClient {
    private var webSocketTask: URLSessionWebSocketTask?
    private var session: URLSession?
    private let encoder = JSONEncoder()

    var onEvent: ((VoxtralEvent) -> Void)?

    func connect(apiKey: String, delayMs: Int) {
        guard let url = URL(string: "wss://api.mistral.ai/v1/audio/transcriptions/realtime") else { return }

        var request = URLRequest(url: url)
        request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")

        session = URLSession(configuration: .default)
        webSocketTask = session?.webSocketTask(with: request)
        webSocketTask?.resume()

        // Send session config
        let config = SessionUpdateMessage(
            session: SessionConfig(
                audioFormat: AudioFormatConfig(),
                targetStreamingDelayMs: delayMs
            )
        )
        sendJSON(config)

        // Start receiving
        receiveLoop()
    }

    func sendAudio(_ pcmData: Data) {
        let base64 = pcmData.base64EncodedString()
        let msg = AudioAppendMessage(audio: base64)
        sendJSON(msg)
    }

    func flush() {
        sendJSON(AudioFlushMessage())
    }

    func disconnect() {
        sendJSON(AudioEndMessage())
        webSocketTask?.cancel(with: .normalClosure, reason: nil)
        webSocketTask = nil
        session?.invalidateAndCancel()
        session = nil
    }

    private func sendJSON<T: Encodable>(_ value: T) {
        guard let data = try? encoder.encode(value),
              let string = String(data: data, encoding: .utf8) else { return }
        webSocketTask?.send(.string(string)) { error in
            if let error {
                print("WebSocket send error: \(error)")
            }
        }
    }

    private func receiveLoop() {
        webSocketTask?.receive { [weak self] result in
            switch result {
            case .success(let message):
                switch message {
                case .string(let text):
                    if let data = text.data(using: .utf8) {
                        let event = parseVoxtralEvent(from: data)
                        Task { @MainActor in
                            self?.onEvent?(event)
                        }
                    }
                case .data(let data):
                    let event = parseVoxtralEvent(from: data)
                    Task { @MainActor in
                        self?.onEvent?(event)
                    }
                @unknown default:
                    break
                }
                self?.receiveLoop()
            case .failure(let error):
                Task { @MainActor in
                    self?.onEvent?(.error("Connection lost: \(error.localizedDescription)"))
                }
            }
        }
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Network/VoxtralWebSocketClient.swift
git commit -m "feat: add Voxtral WebSocket client with connect/send/receive"

---

Task 4: Audio Capture

Files:

• Create: MyVoxtral/MyVoxtral/Audio/AudioCapture.swift

• Step 1: Implement AVAudioEngine mic capture

Create MyVoxtral/MyVoxtral/Audio/AudioCapture.swift:

import AVFoundation

final class AudioCapture {
    private let engine = AVAudioEngine()
    private let targetSampleRate: Double = 16000
    private let chunkDurationMs: Double = 480

    var onChunk: ((Data) -> Void)?

    private var buffer = Data()
    private let bytesPerChunk: Int

    init() {
        // 16kHz * 2 bytes (16-bit) * 1 channel * 0.48s = 15360 bytes
        bytesPerChunk = Int(targetSampleRate * 2 * chunkDurationMs / 1000)
    }

    func start() throws {
        let inputNode = engine.inputNode
        let inputFormat = inputNode.outputFormat(forBus: 0)

        let targetFormat = AVAudioFormat(
            commonFormat: .pcmFormatInt16,
            sampleRate: targetSampleRate,
            channels: 1,
            interleaved: true
        )!

        guard let converter = AVAudioConverter(from: inputFormat, to: targetFormat) else {
            throw AudioCaptureError.converterCreationFailed
        }

        let bufferSize = AVAudioFrameCount(inputFormat.sampleRate * chunkDurationMs / 1000)

        inputNode.installTap(onBus: 0, bufferSize: bufferSize, format: inputFormat) { [weak self] pcmBuffer, _ in
            guard let self else { return }

            let frameCount = AVAudioFrameCount(
                Double(pcmBuffer.frameLength) * self.targetSampleRate / inputFormat.sampleRate
            )
            guard let convertedBuffer = AVAudioPCMBuffer(
                pcmFormat: targetFormat,
                frameCapacity: frameCount
            ) else { return }

            var error: NSError?
            let status = converter.convert(to: convertedBuffer, error: &error) { _, outStatus in
                outStatus.pointee = .haveData
                return pcmBuffer
            }

            guard status != .error, error == nil else { return }

            let byteCount = Int(convertedBuffer.frameLength) * 2 // 16-bit = 2 bytes
            guard let int16Ptr = convertedBuffer.int16ChannelData?[0] else { return }
            let data = Data(bytes: int16Ptr, count: byteCount)

            self.buffer.append(data)

            while self.buffer.count >= self.bytesPerChunk {
                let chunk = self.buffer.prefix(self.bytesPerChunk)
                self.buffer = Data(self.buffer.dropFirst(self.bytesPerChunk))
                self.onChunk?(Data(chunk))
            }
        }

        engine.prepare()
        try engine.start()
    }

    func stop() {
        engine.inputNode.removeTap(onBus: 0)
        engine.stop()

        // Flush remaining buffer
        if !buffer.isEmpty {
            onChunk?(buffer)
            buffer = Data()
        }
    }
}

enum AudioCaptureError: Error, LocalizedError {
    case converterCreationFailed

    var errorDescription: String? {
        switch self {
        case .converterCreationFailed:
            return "Failed to create audio format converter"
        }
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Audio/AudioCapture.swift
git commit -m "feat: add AudioCapture with AVAudioEngine mic tap and PCM conversion"

---

Task 5: App Settings

Files:

• Create: MyVoxtral/MyVoxtral/Models/AppSettings.swift

• Step 1: Implement settings model

Create MyVoxtral/MyVoxtral/Models/AppSettings.swift:

import SwiftUI
import Carbon.HIToolbox

enum OutputMode: String, CaseIterable {
    case textBox = "Text Window"
    case cursorInjection = "Type at Cursor"
}

final class AppSettings: ObservableObject {
    static let shared = AppSettings()

    @AppStorage("apiKey") var apiKey: String = ""
    @AppStorage("outputMode") var outputMode: OutputMode = .textBox
    @AppStorage("streamingDelayMs") var streamingDelayMs: Int = 480
    @AppStorage("shortcutKeyCode") var shortcutKeyCode: UInt16 = 0
    @AppStorage("shortcutModifiers") var shortcutModifiers: UInt = 0

    var hasAPIKey: Bool { !apiKey.isEmpty }

    var hasShortcut: Bool { shortcutKeyCode != 0 || shortcutModifiers != 0 }

    var shortcutDisplayString: String {
        guard hasShortcut else { return "Not Set" }
        var parts: [String] = []
        let mods = NSEvent.ModifierFlags(rawValue: shortcutModifiers)
        if mods.contains(.control) { parts.append("^") }
        if mods.contains(.option) { parts.append("\u{2325}") }
        if mods.contains(.shift) { parts.append("\u{21E7}") }
        if mods.contains(.command) { parts.append("\u{2318}") }
        // Map key code to character
        if let scalar = Unicode.Scalar(shortcutKeyCode) {
            parts.append(String(Character(scalar)).uppercased())
        }
        return parts.joined()
    }
}

extension OutputMode: RawRepresentable where RawValue == String {}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Models/AppSettings.swift
git commit -m "feat: add AppSettings with API key, shortcut, output mode, delay"

---

Task 6: Transcription Logger

Files:

• Create: MyVoxtral/MyVoxtral/Utilities/TranscriptionLogger.swift

• Step 1: Implement the log file writer

Create MyVoxtral/MyVoxtral/Utilities/TranscriptionLogger.swift:

import Foundation

struct TranscriptionLogger {
    private static var logFileURL: URL {
        let appSupport = FileManager.default.urls(for: .applicationSupportDirectory, in: .userDomainMask).first!
        let dir = appSupport.appendingPathComponent("MyVoxtral", isDirectory: true)
        try? FileManager.default.createDirectory(at: dir, withIntermediateDirectories: true)
        return dir.appendingPathComponent("transcription.log")
    }

    static func append(text: String) {
        let timestamp = ISO8601DateFormatter().string(from: Date())
        let entry = "[\(timestamp)]\n\(text)\n---\n\n"

        guard let data = entry.data(using: .utf8) else { return }

        if FileManager.default.fileExists(atPath: logFileURL.path) {
            if let handle = try? FileHandle(forWritingTo: logFileURL) {
                handle.seekToEndOfFile()
                handle.write(data)
                handle.closeFile()
            }
        } else {
            try? data.write(to: logFileURL)
        }
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Utilities/TranscriptionLogger.swift
git commit -m "feat: add TranscriptionLogger for append-only session log"

---

Task 7: Cursor Injector

Files:

• Create: MyVoxtral/MyVoxtral/Utilities/CursorInjector.swift

• Step 1: Implement CGEvent keystroke simulation

Create MyVoxtral/MyVoxtral/Utilities/CursorInjector.swift:

import ApplicationServices

struct CursorInjector {
    static var isAccessibilityGranted: Bool {
        AXIsProcessTrusted()
    }

    static func promptAccessibilityPermission() {
        let options = [kAXTrustedCheckOptionPrompt.takeUnretainedValue() as String: true] as CFDictionary
        AXIsProcessTrustedWithOptions(options)
    }

    static func typeText(_ text: String) {
        guard isAccessibilityGranted else { return }

        let source = CGEventSource(stateID: .hidSystemState)

        for character in text {
            let string = String(character)
            let event = CGEvent(keyboardEventSource: source, virtualKey: 0, keyDown: true)
            event?.keyboardSetUnicodeString(stringLength: string.utf16.count, unicodeString: Array(string.utf16))
            event?.post(tap: .cghidEventTap)

            let eventUp = CGEvent(keyboardEventSource: source, virtualKey: 0, keyDown: false)
            eventUp?.keyboardSetUnicodeString(stringLength: string.utf16.count, unicodeString: Array(string.utf16))
            eventUp?.post(tap: .cghidEventTap)
        }
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Utilities/CursorInjector.swift
git commit -m "feat: add CursorInjector with CGEvent keystroke simulation"

---

Task 8: Global Shortcut

Files:

• Create: MyVoxtral/MyVoxtral/Utilities/GlobalShortcut.swift

• Step 1: Implement global keyboard shortcut monitor

Create MyVoxtral/MyVoxtral/Utilities/GlobalShortcut.swift:

import Cocoa

final class GlobalShortcut {
    private var monitor: Any?
    var onTrigger: (() -> Void)?

    func register(keyCode: UInt16, modifiers: UInt) {
        unregister()
        guard keyCode != 0 || modifiers != 0 else { return }

        let requiredFlags = NSEvent.ModifierFlags(rawValue: modifiers)

        monitor = NSEvent.addGlobalMonitorForEvents(matching: .keyDown) { [weak self] event in
            let mask: NSEvent.ModifierFlags = [.command, .option, .control, .shift]
            if event.keyCode == keyCode && event.modifierFlags.intersection(mask) == requiredFlags {
                self?.onTrigger?()
            }
        }
    }

    func unregister() {
        if let monitor {
            NSEvent.removeMonitor(monitor)
        }
        monitor = nil
    }

    deinit {
        unregister()
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Utilities/GlobalShortcut.swift
git commit -m "feat: add configurable global keyboard shortcut"

---

Task 9: Transcription Manager

Files:

• Create: MyVoxtral/MyVoxtral/Models/TranscriptionManager.swift

• Step 1: Implement the orchestrator

Create MyVoxtral/MyVoxtral/Models/TranscriptionManager.swift:

import SwiftUI

enum RecordingState: Equatable {
    case idle
    case recording
    case error(String)
}

@MainActor
final class TranscriptionManager: ObservableObject {
    @Published var state: RecordingState = .idle
    @Published var currentText: String = ""

    private let audioCapture = AudioCapture()
    private let wsClient = VoxtralWebSocketClient()
    private let settings = AppSettings.shared
    private var hasRetried = false

    var isRecording: Bool { state == .recording }

    func toggle() {
        if isRecording {
            stop()
        } else {
            start()
        }
    }

    func start() {
        guard settings.hasAPIKey else {
            state = .error("No API key set. Open Settings.")
            return
        }

        currentText = ""
        hasRetried = false

        wsClient.onEvent = { [weak self] event in
            self?.handleEvent(event)
        }

        wsClient.connect(apiKey: settings.apiKey, delayMs: settings.streamingDelayMs)

        audioCapture.onChunk = { [weak self] chunk in
            Task { @MainActor in
                self?.wsClient.sendAudio(chunk)
            }
        }

        do {
            try audioCapture.start()
            state = .recording
        } catch {
            state = .error("Mic error: \(error.localizedDescription)")
        }
    }

    func stop() {
        audioCapture.stop()
        wsClient.flush()
        wsClient.disconnect()
        state = .idle

        if !currentText.isEmpty {
            TranscriptionLogger.append(text: currentText)
        }
    }

    private func handleEvent(_ event: VoxtralEvent) {
        switch event {
        case .sessionCreated:
            break
        case .textDelta(let text):
            currentText += text
            if settings.outputMode == .cursorInjection {
                CursorInjector.typeText(text)
            }
        case .segment:
            break // text deltas already handle text accumulation
        case .language:
            break
        case .done(let text):
            if currentText.isEmpty {
                currentText = text
            }
        case .error(let message):
            if !hasRetried && state == .recording {
                hasRetried = true
                wsClient.disconnect()
                wsClient.connect(apiKey: settings.apiKey, delayMs: settings.streamingDelayMs)
            } else {
                state = .error(message)
                audioCapture.stop()
            }
        case .unknown:
            break
        }
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Models/TranscriptionManager.swift
git commit -m "feat: add TranscriptionManager orchestrating audio, WS, and output"

---

Task 10: Floating Transcription Window

Files:

• Create: MyVoxtral/MyVoxtral/Views/TranscriptionWindow.swift

• Step 1: Implement NSPanel-based floating window

Create MyVoxtral/MyVoxtral/Views/TranscriptionWindow.swift:

import SwiftUI
import AppKit

struct TranscriptionContentView: View {
    @ObservedObject var manager: TranscriptionManager

    var body: some View {
        VStack(alignment: .leading, spacing: 8) {
            HStack {
                Circle()
                    .fill(manager.isRecording ? .red : .gray)
                    .frame(width: 8, height: 8)
                Text(manager.isRecording ? "Recording..." : "Idle")
                    .font(.caption)
                    .foregroundStyle(.secondary)
                Spacer()
                Button {
                    NSPasteboard.general.clearContents()
                    NSPasteboard.general.setString(manager.currentText, forType: .string)
                } label: {
                    Image(systemName: "doc.on.doc")
                }
                .buttonStyle(.borderless)
                .disabled(manager.currentText.isEmpty)
            }

            ScrollViewReader { proxy in
                ScrollView {
                    Text(manager.currentText.isEmpty ? "Transcription will appear here..." : manager.currentText)
                        .frame(maxWidth: .infinity, alignment: .leading)
                        .foregroundStyle(manager.currentText.isEmpty ? .secondary : .primary)
                        .textSelection(.enabled)
                        .id("bottom")
                }
                .onChange(of: manager.currentText) {
                    proxy.scrollTo("bottom", anchor: .bottom)
                }
            }
        }
        .padding()
        .frame(width: 320, height: 200)
    }
}

final class TranscriptionPanel {
    private var panel: NSPanel?
    private let manager: TranscriptionManager

    init(manager: TranscriptionManager) {
        self.manager = manager
    }

    func show() {
        if panel == nil {
            let panel = NSPanel(
                contentRect: NSRect(x: 0, y: 0, width: 320, height: 200),
                styleMask: [.titled, .closable, .resizable, .nonactivatingPanel, .utilityWindow],
                backing: .buffered,
                defer: false
            )
            panel.title = "MyVoxtral"
            panel.isFloatingPanel = true
            panel.level = .floating
            panel.contentView = NSHostingView(rootView: TranscriptionContentView(manager: manager))
            panel.center()
            self.panel = panel
        }
        panel?.orderFront(nil)
    }

    func hide() {
        panel?.orderOut(nil)
    }

    var isVisible: Bool {
        panel?.isVisible ?? false
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Views/TranscriptionWindow.swift
git commit -m "feat: add floating transcription panel with auto-scroll and copy"

---

Task 11: Settings View

Files:

• Create: MyVoxtral/MyVoxtral/Views/SettingsView.swift

• Step 1: Implement settings window

Create MyVoxtral/MyVoxtral/Views/SettingsView.swift:

import SwiftUI

struct SettingsView: View {
    @ObservedObject var settings = AppSettings.shared
    @State private var isRecordingShortcut = false

    var body: some View {
        Form {
            Section("API") {
                SecureField("Mistral API Key", text: $settings.apiKey)
            }

            Section("Output") {
                Picker("Mode", selection: $settings.outputMode) {
                    ForEach(OutputMode.allCases, id: \.self) { mode in
                        Text(mode.rawValue).tag(mode)
                    }
                }
                .pickerStyle(.segmented)

                if settings.outputMode == .cursorInjection && !CursorInjector.isAccessibilityGranted {
                    HStack {
                        Image(systemName: "exclamationmark.triangle.fill")
                            .foregroundStyle(.yellow)
                        Text("Accessibility permission required")
                            .font(.caption)
                        Button("Grant") {
                            CursorInjector.promptAccessibilityPermission()
                        }
                        .font(.caption)
                    }
                }
            }

            Section("Shortcut") {
                HStack {
                    Text("Toggle Recording:")
                    Spacer()
                    Button(isRecordingShortcut ? "Press keys..." : settings.shortcutDisplayString) {
                        isRecordingShortcut = true
                    }
                    .onKeyPress { press in
                        guard isRecordingShortcut else { return .ignored }
                        settings.shortcutKeyCode = press.key.character.flatMap {
                            UInt16($0.asciiValue ?? 0)
                        } ?? 0
                        settings.shortcutModifiers = press.modifiers.rawValue
                        isRecordingShortcut = false
                        return .handled
                    }
                }
            }

            Section("Latency") {
                VStack(alignment: .leading) {
                    Text("Streaming delay: \(settings.streamingDelayMs)ms")
                        .font(.caption)
                    Slider(
                        value: Binding(
                            get: { Double(settings.streamingDelayMs) },
                            set: { settings.streamingDelayMs = Int($0) }
                        ),
                        in: 240...2400,
                        step: 120
                    )
                    HStack {
                        Text("Fast").font(.caption2).foregroundStyle(.secondary)
                        Spacer()
                        Text("Accurate").font(.caption2).foregroundStyle(.secondary)
                    }
                }
            }
        }
        .formStyle(.grouped)
        .frame(width: 360, height: 340)
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Views/SettingsView.swift
git commit -m "feat: add SettingsView with API key, shortcut, mode, and latency"

---

Task 12: Menu Bar View

Files:

• Create: MyVoxtral/MyVoxtral/Views/MenuBarView.swift

• Step 1: Implement the menu bar dropdown

Create MyVoxtral/MyVoxtral/Views/MenuBarView.swift:

import SwiftUI

struct MenuBarView: View {
    @ObservedObject var manager: TranscriptionManager
    @ObservedObject var settings = AppSettings.shared
    let onShowTranscription: () -> Void
    let onShowSettings: () -> Void

    var body: some View {
        VStack(spacing: 4) {
            Button(manager.isRecording ? "Stop Recording" : "Start Recording") {
                manager.toggle()
            }
            .keyboardShortcut("r")

            if case .error(let msg) = manager.state {
                Text(msg)
                    .font(.caption)
                    .foregroundStyle(.red)
                    .lineLimit(2)
                    .padding(.horizontal, 8)
            }

            Divider()

            Button("Show Transcription") {
                onShowTranscription()
            }

            Button("Settings...") {
                onShowSettings()
            }

            Divider()

            Button("Quit") {
                NSApplication.shared.terminate(nil)
            }
            .keyboardShortcut("q")
        }
        .padding(.vertical, 4)
    }
}

• Step 2: Build to verify

swift build

Expected: Build succeeds.

• Step 3: Commit

git add MyVoxtral/Views/MenuBarView.swift
git commit -m "feat: add MenuBarView with start/stop, settings, and error display"

---

Task 13: Wire Everything Together in App Entry

Files:

• Modify: MyVoxtral/MyVoxtral/MyVoxtralApp.swift

• Step 1: Update the app entry point to integrate all components

Replace the contents of MyVoxtral/MyVoxtral/MyVoxtralApp.swift with:

import SwiftUI

@main
struct MyVoxtralApp: App {
    @StateObject private var manager = TranscriptionManager()
    @StateObject private var settings = AppSettings.shared
    @State private var transcriptionPanel: TranscriptionPanel?
    @State private var settingsWindow: NSWindow?

    private let globalShortcut = GlobalShortcut()

    var body: some Scene {
        MenuBarExtra {
            MenuBarView(
                manager: manager,
                onShowTranscription: { showTranscriptionWindow() },
                onShowSettings: { showSettingsWindow() }
            )
        } label: {
            Image(systemName: manager.isRecording ? "mic.fill" : "mic")
                .symbolRenderingMode(.palette)
                .foregroundStyle(manager.isRecording ? .red : .primary)
        }
        .onChange(of: settings.shortcutKeyCode) { registerShortcut() }
        .onChange(of: settings.shortcutModifiers) { registerShortcut() }
        .onAppear {
            if !settings.hasAPIKey {
                showSettingsWindow()
            }
            registerShortcut()
        }
        .onChange(of: manager.isRecording) {
            if manager.isRecording && settings.outputMode == .textBox {
                showTranscriptionWindow()
            }
        }
    }

    private func registerShortcut() {
        globalShortcut.register(
            keyCode: settings.shortcutKeyCode,
            modifiers: settings.shortcutModifiers
        )
        globalShortcut.onTrigger = { [weak manager] in
            Task { @MainActor in
                manager?.toggle()
            }
        }
    }

    private func showTranscriptionWindow() {
        if transcriptionPanel == nil {
            transcriptionPanel = TranscriptionPanel(manager: manager)
        }
        transcriptionPanel?.show()
    }

    private func showSettingsWindow() {
        if let settingsWindow, settingsWindow.isVisible {
            settingsWindow.makeKeyAndOrderFront(nil)
            return
        }
        let window = NSWindow(
            contentRect: NSRect(x: 0, y: 0, width: 360, height: 340),
            styleMask: [.titled, .closable],
            backing: .buffered,
            defer: false
        )
        window.title = "MyVoxtral Settings"
        window.contentView = NSHostingView(rootView: SettingsView())
        window.center()
        window.makeKeyAndOrderFront(nil)
        NSApp.activate(ignoringOtherApps: true)
        self.settingsWindow = window
    }
}

• Step 2: Build the full app

swift build

Expected: Build succeeds with all components linked.

• Step 3: Run and verify basic functionality

swift run

Expected: Menu bar icon appears. Clicking shows the dropdown with Start/Stop, Settings, Quit. Settings window opens if no API key.

• Step 4: Commit

git add MyVoxtral/MyVoxtralApp.swift
git commit -m "feat: wire all components into app entry with menu bar, shortcut, and windows"

---

Task 14: End-to-End Smoke Test

• Step 1: Verify the full recording flow

swift run

Manual test checklist:

1. App launches as menu bar icon (no Dock icon)
2. Click icon → dropdown shows Start Recording, Settings, Quit
3. Open Settings → enter a valid Mistral API key
4. Set output mode to "Text Window"
5. Click "Start Recording" → mic icon turns red, transcription window opens
6. Speak into microphone → text appears in the floating window
7. Click "Stop Recording" → icon returns to normal
8. Verify ~/Library/Application Support/MyVoxtral/transcription.log contains the session
9. Switch to "Type at Cursor" mode → grant Accessibility permission if prompted
10. Open TextEdit, click Start Recording, speak → text appears at cursor in TextEdit
11. Configure a keyboard shortcut in Settings → verify it toggles recording from any app

• Step 2: Fix any issues found during smoke testing

Address any build errors, runtime crashes, or connection issues.

• Step 3: Final commit

git add -A
git commit -m "test: verify end-to-end transcription flow"