← Back to the Suite
Voice Interface

Talk to Claude.
Hands-free.

Voice interface for Claude Code. Speak to Claude instead of typing — in any macOS app, on any task, without lifting your hands from your side.

Get Started Free →

Floating orb.
Menu bar popover.

Two surfaces, always within reach. The orb lives wherever you drag it. The popover gives you full control.

Floating Indicator
Claude
Recording
Idle
Processing
Speaking

Drag anywhere.
Stays above all windows.

Menu Bar Popover
AgentWhisper
Speaking
Mode:
Dictation
Claude
Start Recording
You said:
Update the auth middleware to use JWT tokens
Claude:
Stop Clear
I'll update the auth middleware.
Reading src/middleware/auth.ts...

Replacing session auth with
JWT verification + refresh...
TTS
Hub

Talk to your agents.

AgentWhisper is a macOS menu bar app that converts speech to text and injects it into any application — with direct Claude Code integration built in from the start.

🎯
Universal Dictation

Speak into any macOS app — editors, chat, email, browsers, terminals. Hold the hotkey, talk, release. Your words appear as typed text instantly, wherever your cursor is.

⌨️
Flexible Hotkey Modes

Three hotkey modes to match your style: Key Combination (modifier + key, single press to toggle), Hold/Push-to-Talk (default Right Option — hold to record, release to transcribe), or Smart (tap under 200ms to toggle, hold 200ms or more for PTT). Fully customizable.

🤖
Claude Code Integration

Speak a prompt and AgentWhisper spawns a claude -p subprocess, streams its output to the popover, and reads the response back aloud. True hands-free pair programming — no plugin or extension required.

🎤
5 STT Providers

Choose your transcription engine: Apple Speech (free, fully on-device), whisper-cpp (local, offline, no API key), OpenAI Whisper ($0.006/min), Groq Whisper Large v3 ($0.005/min, fastest cloud option), or Deepgram Nova-2 ($0.0043/min). Trade off privacy, cost, and accuracy.

🖥️
Power Mode

App-based profiles using bundle ID detection. Each profile sets its own dictation mode, speech provider, and auto-enter behavior. Ships with defaults for Terminal, Code Editor, and Chat/Email — so switching apps automatically switches your voice setup.

🔊
Voice Chat & TTS

Direct voice conversation with Claude via the Anthropic API — no terminal required. Responses are read back using macOS system voices (20+ options via AVSpeechSynthesizer, 150–450 WPM configurable) or ElevenLabs eleven_multilingual_v2 for premium voices.

🔒
100% Local Processing Available

Choose Apple Speech as your transcription provider and your voice never leaves your Mac. No audio sent to any server, no cloud dependency — on-device processing via Apple's SFSpeechRecognizer. Your voice stays yours.

Native SwiftUI.
Lives in your menu bar.

AgentWhisper is a lightweight macOS menu bar app — always available, never in the way, and completely private when you want it to be.

Platform & STT Providers
  • macOS 13.0+ (Ventura and later) — Apple Silicon (M2 or later recommended)
  • Native SwiftUI menu bar app — no Electron
  • Apple Speech — free, fully on-device (SFSpeechRecognizer, sessions roll every 55s)
  • whisper-cpp — local, offline, no API key required
  • OpenAI Whisper — cloud, whisper-1, $0.006/min
  • Groq Whisper Large v3 — cloud, $0.005/min (fastest)
  • Deepgram Nova-2 — cloud, $0.0043/min
TTS, Modes & Integrations
  • System voices — macOS say / AVSpeechSynthesizer, 20+ voices, 150–450 WPM
  • ElevenLabs — eleven_multilingual_v2, API key stored in Keychain
  • Dictation modes: Inject, Claude subprocess, Voice Chat
  • Hotkey modes: Key Combo, Hold/PTT, Smart (tap vs hold)
  • Power Mode — per-app profiles via bundle ID
  • Claude Code (spawns claude -p subprocess, streams output)
  • QuantaCanvas PTY backchannel for terminal inject

Part of the AgentQuanta Suite

AgentWhisper is the voice layer. Pair it with AgentTUI to run agents hands-free, and AgentEKG to monitor every prompt you speak.