Configuration

All configuration is done through the Settings modal in the app.

Settings Modal

The Settings modal opens automatically on first launch. Reopen it anytime by clicking the gear icon in the main screen.

Gateway Settings

Setting	Description	Default
Gateway URL	WebSocket URL for the OpenClaw gateway	ws://127.0.0.1:18789
GATEWAY_TOKEN	Authentication token for the gateway	—

The gateway connects to an OpenClaw instance that routes requests to your configured LLM.

STT Settings

Setting	Description	Default
STT Provider	Speech-to-text engine	ElevenLabs Scribe
ELEVENLABS_API_KEY	API key (ElevenLabs)	—
OPENAI_API_KEY	API key (OpenAI Whisper)	—
whisper.cpp Binary Path	Path to whisper-cli binary	—

See Providers for details on each provider.

TTS Settings

Setting	Description	Default
TTS Provider	Text-to-speech engine	ElevenLabs
TTS Voice ID	ElevenLabs voice	pFZP5JQG7iQjIQuC4Bku (Lily)
TTS Model	ElevenLabs model	eleven_multilingual_v2
VOICEVOX URL	VOICEVOX server URL	http://localhost:50021
VOICEVOX Speaker ID	VOICEVOX speaker	1
Kokoro URL	Kokoro server URL	http://localhost:8880
Kokoro Voice	Kokoro voice ID	jf_alpha
Piper Binary Path	Path to piper binary	—
Piper Model Path	Path to .onnx model	—

See Providers for details on each provider.

VAD Settings

Setting	Description	Default
Sensitivity	auto or manual value (0.001–0.05)	auto

See Voice Activity Detection below.

Connectivity Checks

Each section has a Test button that validates the connection:

• Gateway: HTTP GET to the gateway URL (requires GATEWAY_TOKEN)
• STT: Provider-specific check (API key validation or binary existence)
• TTS: Provider-specific check (API key validation, server connectivity, or binary existence)

All three checks must pass before the Start button is enabled.

---

API Key Management

Storage

API keys are encrypted with AES-256-CBC and stored locally:

Platform	Location
macOS	~/Library/Application Support/TalkLob/keys.json
Windows	%APPDATA%/TalkLob/keys.json
Linux	~/.config/TalkLob/keys.json

Keys never leave your machine unencrypted.

Key Sources

Each key can be loaded from three sources (shown as buttons in the Settings modal):

Source	Description
Manual	Type the key directly into the input field
OpenClaw	Auto-load from ~/.openclaw/openclaw.json
Env	Read from environment variables

OpenClaw Config Format

{
  "gateway": {
    "auth": { "token": "your-gateway-token" }
  },
  "env": {
    "ELEVENLABS_API_KEY": "your-elevenlabs-key",
    "OPENAI_API_KEY": "your-openai-key"
  }
}

Environment Variables

Variable	Used For
GATEWAY_TOKEN	OpenClaw gateway authentication
ELEVENLABS_API_KEY	ElevenLabs STT and TTS
OPENAI_API_KEY	OpenAI Whisper STT

---

Voice Activity Detection

TalkLob uses Silero VAD (neural network-based) to detect when you start and stop speaking. No push-to-talk button needed.

Auto Calibration (Default)

When you enable the microphone, the app runs a 1.5-second calibration:

1. Captures ambient noise via the microphone (with WebRTC echo cancellation enabled)
2. Measures the median RMS (root mean square) noise level
3. Maps the noise floor to a VAD threshold:
   • Quiet room (low RMS) → lower threshold → more sensitive
   • Noisy environment (high RMS) → higher threshold → less sensitive

During calibration, the status shows "Calibrating...".

Manual Sensitivity

If auto calibration doesn't work well for your environment, switch to manual mode in VAD Settings and adjust the slider. Lower values = more sensitive (picks up quieter speech), higher values = less sensitive (ignores more background noise).

---

Speaker Monitor

The speaker monitor captures system audio output to detect when media (YouTube, music, etc.) is playing. When system audio is detected, VAD is temporarily suppressed to prevent the AI from responding to non-speech sounds.

• Uses Electron's desktopCapturer API
• 800ms debounce to avoid rapid toggling
• Fails gracefully if system audio capture is not available

---

Aizuchi (Backchanneling)

During the Thinking state (while waiting for the LLM response), TalkLob plays subtle audio cues to fill the silence and signal that the AI is processing. This mimics the Japanese conversational habit of "aizuchi" (相槌).

• Initial delay: 1.5–2.5 seconds
• Interval: 3–5 seconds between cues
• Automatically stops when the AI starts speaking

---

Settings File

Settings are stored as JSON at ~/.config/lobster/settings.json. While you can edit this file directly, using the Settings modal is recommended.

{
  "gatewayUrl": "ws://127.0.0.1:18789",
  "sttProvider": "elevenlabs",
  "ttsProvider": "elevenlabs",
  "ttsVoiceId": "pFZP5JQG7iQjIQuC4Bku",
  "ttsModelId": "eleven_multilingual_v2",
  "voicevoxUrl": "http://localhost:50021",
  "voicevoxSpeakerId": 1,
  "kokoroUrl": "http://localhost:8880",
  "kokoroVoice": "jf_alpha",
  "vadSensitivity": "auto"
}