Voice & Text-to-Speech Setup

## Fix MEDIA: Path Appearing as Text When you see output like: `` 🦞 Got it. MEDIA: /Users/you/.openclaw/cache/tts/abc123.ogg `` Instead of receiving an actual voice note, try these fixes:

### 1. Verify TTS Configuration Check your current TTS settings:

openclaw config get messages.tts

You should see something like: ``json5 { "provider": "elevenlabs", "auto": "always", "elevenlabs": { "voiceId": "FF7KdobWPaiR0vkcALHF", "modelId": "eleven_v3" } }

### 2. Ensure Provider API Key is Set ElevenLabs requires a valid API key:

openclaw config set messages.tts.elevenlabs.apiKey "your-elevenlabs-api-key"

Get your API key from: https://elevenlabs.io/app/settings/api-keys

### 3. Restart the Gateway After any TTS configuration change, restart:

openclaw gateway restart

### 4. Check Channel Voice Capability Not all channels support voice notes the same way. Verify your channel: - Telegram: Full voice note support via asVoice: true - Discord: Voice attachments work as audio files - Slack/iMessage: May render as file attachments

### 5. Test TTS Directly Ask the agent explicitly to speak: `` Say "hello world" as a voice note Or use the /tts command if available: /tts Hello, this is a test

### 6. Check for Model Instruction Issues Some models may not properly handle TTS tool calls. If using a non-Anthropic model (like Kimi K2.5), ensure it understands the TTS tool schema. You may need to add explicit instructions in your IDENTITY.md: ``markdown ## Voice Output When TTS is enabled, use the tts tool to convert responses to speech. Do not output MEDIA: paths as text - the system handles audio attachment automatically.

--- ## Hands-Free Voice-Only Setup (Car/Bluetooth) For safe driving use with Bluetooth, configure full voice loop:

### 1. Enable Always-On TTS Make the agent always respond with voice:

openclaw config set messages.tts.auto "always"openclaw config set messages.tts.provider "elevenlabs"

### 2. Configure Speech-to-Text (STT) Enable STT so you can speak instead of typing:

openclaw config set messages.stt.provider "elevenlabs"openclaw config set messages.stt.auto true

Alternative STT providers: - whisper - OpenAI Whisper (local or API) - deepgram - Fast and accurate - google - Google Cloud Speech

### 3. Telegram Voice Message Workflow For Telegram, the safest hands-free flow: 1. Send voice messages - Telegram converts your speech to a voice note 2. OpenClaw transcribes - STT converts your voice to text 3. Agent processes - Responds to your request 4. TTS converts - Response becomes a voice note 5. Bluetooth plays - You hear the response through car speakers

### 4. iOS Shortcuts Integration (Advanced) Create an iOS Shortcut for truly hands-free activation: 1. Create a Shortcut that sends a message to your Telegram bot 2. Use "Hey Siri" to trigger the shortcut 3. Dictate your message 4. The agent responds via voice note Example Shortcut actions: - Get input from Spoken Text - Send message to Telegram chat - Wait for response - Play audio

### 5. Safety Configuration For driving, add guardrails in your IDENTITY.md: ``markdown ## Driving Mode Keep responses brief and clear for audio playback. No visual elements (tables, code blocks) - voice-friendly only. Confirm critical actions verbally before executing.