Voice Configuration

    Wingman supports text-to-speech using Web Speech (browser-based) or ElevenLabs.

    Global voice config

    {
  "voice": {
    "provider": "web_speech",
    "defaultPolicy": "manual",
    "webSpeech": {
      "voiceName": "Samantha",
      "lang": "en-US",
      "rate": 1,
      "pitch": 1,
      "volume": 0.9
    }
  }
}

    Providers

    • web_speech: uses the browser's SpeechSynthesis API (Control UI only).
    • elevenlabs: uses ElevenLabs TTS API (gateway-side).

    ElevenLabs example

    {
  "voice": {
    "provider": "elevenlabs",
    "defaultPolicy": "auto",
    "elevenlabs": {
      "voiceId": "voice_123",
      "modelId": "eleven_turbo_v2",
      "stability": 0.45,
      "similarityBoost": 0.7,
      "style": 0.2,
      "speakerBoost": true,
      "speed": 1,
      "outputFormat": "mp3_44100_128",
      "optimizeStreamingLatency": 2
    }
  }
}

    Set your API key via:

    export ELEVENLABS_API_KEY="..."

    Per-agent overrides

    {
  "name": "narrator",
  "description": "Reads results aloud.",
  "systemPrompt": "Summarize and narrate results.",
  "voice": {
    "provider": "elevenlabs",
    "elevenlabs": { "voiceId": "voice_abc" }
  }
}

    Voice policy

    defaultPolicy controls when speech triggers:

    • off: disabled
    • manual: only when requested
    • auto: automatic speech for responses

    How it works

    • The Control UI reads /api/voice for current settings.
    • /api/voice/speak streams audio when using ElevenLabs.
    • Web Speech runs entirely in the browser, so it does not require gateway TTS access.