ClawPilot - Discord Voice for OpenClaw

Talk to your OpenClaw agent through Discord voice channels.

Setup — Step by step

Step 1: Create a Discord Bot

1. Go to https://discord.com/developers/applications

2. Click "New Application" — give it a name (e.g. "ClawPilot")

3. Go to the Bot section (left sidebar)

4. Click "Reset Token" to generate a bot token — copy it, you'll need it later

5. Under Privileged Gateway Intents — no privileged intents are needed. Leave everything off.

6. Go to OAuth2 > URL Generator (left sidebar)

7. Under Scopes, check: bot, applications.commands

8. Under Bot Permissions, check exactly these 5 permissions:

   General Permissions:

   • View Channels

   Text Permissions:

   • Send Messages
   • Use Slash Commands

   Voice Permissions:

   • Connect
   • Speak

9. Copy the generated URL at the bottom and open it in your browser

10. Select your Discord server and click Authorize

Step 2: Choose your settings

What language will you mainly speak?

• fr — French
• en — English
• es — Spanish
• de — German
• Any ISO language code

What name should the agent respond to?
Pick a short, easy-to-pronounce name (e.g. "bobby", "claw", "jarvis", "friday"). You'll say this name in voice to activate the agent.

Which STT (Speech-to-Text) provider?

Provider	Latency	Cost	Setup
whisper-local	~2-3s	Free	Install whisper.cpp + download a model
deepgram	~200ms	~$0.05/h	Get API key at deepgram.com
whisper	~2-3s	~$0.006/min	Get API key at platform.openai.com

Which TTS (Text-to-Speech) provider?

Provider	Latency	Cost	Setup
edge	~1-2s	Free	pip install edge-tts + ffmpeg
openai	~300ms	~$15/1M chars	Get API key at platform.openai.com

Which TTS voice?

• For Edge TTS (free), pick from the voice list. Popular choices:
  • French: fr-FR-RemyMultilingualNeural (warm male), fr-FR-DeniseNeural (female)
  • English: en-US-AndrewMultilingualNeural (natural male), en-US-AriaNeural (female)
• For OpenAI TTS: nova (female), onyx (male), alloy (neutral), echo, fable, shimmer

Which model for the voice agent?
The voice agent should use a fast model for quick responses:

• anthropic/claude-sonnet-4-5-20250929 — recommended (fast + smart)
• anthropic/claude-haiku-3-5-20241022 — fastest, cheapest, good for simple Q&A

Step 3: Install ClawPilot

git clone https://github.com/CryptoManiaques/ClawPilot.git
cd ClawPilot
npm install

For whisper-local STT (free):

# macOS
brew install whisper-cpp ffmpeg
pip install edge-tts
whisper-cpp-download-ggml-model tiny  # or: base, small, medium

# Linux (Ubuntu/Debian)
apt install ffmpeg
pip install edge-tts
# Build whisper.cpp from source: https://github.com/ggerganov/whisper.cpp

Register as an OpenClaw plugin:

openclaw plugins install --link /path/to/ClawPilot

Step 4: Create a voice agent

Add a dedicated voice agent to your OpenClaw config (~/.openclaw/openclaw.json). This agent uses a fast model optimized for short voice responses:

{
  "agents": {
    "list": [
      // ... your existing agents ...
      {
        "id": "voice",
        "name": "YOUR_AGENT_NAME Voice",
        "model": {
          "primary": "anthropic/claude-sonnet-4-5-20250929"
        },
        "workspace": "~/.openclaw/agents/voice",
        "agentDir": "~/.openclaw/agents/voice/agent"
      }
    ]
  }
}

Create the voice agent system prompt:

mkdir -p ~/.openclaw/agents/voice/agent

Write ~/.openclaw/agents/voice/agent/AGENT.md:

# Voice Agent — YOUR_AGENT_NAME

You are YOUR_AGENT_NAME, a friendly AI assistant responding via Discord voice chat.
Your responses are converted to speech, so write exactly as you would speak.

## Rules (STRICT)
- Maximum 1-2 sentences. NEVER more than 3 sentences.
- Be conversational and warm, like a friend
- Match the user's language (French or English)
- NEVER use markdown: no asterisks, no bold, no italic, no headers, no lists, no code blocks
- NEVER use emojis or emoticons
- Write pure plain spoken text only
- If unsure, ask to repeat briefly
- Answer directly, no filler phrases

Step 5: Configure the plugin

Add to your OpenClaw config under plugins.entries:

{
  "plugins": {
    "entries": {
      "clawpilot": {
        "enabled": true,
        "config": {
          "discordToken": "YOUR_DISCORD_BOT_TOKEN",
          "sttProvider": "whisper-local",
          "ttsProvider": "edge",
          "edgeTtsVoice": "fr-FR-RemyMultilingualNeural",
          "agentName": "bobby",
          "agentId": "voice",
          "activationMode": "always_active"
        }
      }
    }
  }
}

Replace agentName with the name you chose, edgeTtsVoice with your preferred voice, and agentId with "voice" to use the dedicated fast voice agent.

Step 6: Start

openclaw gateway restart

Then in Discord:

1. Join a voice channel
2. Type /join in any text channel
3. Speak! Say your agent's name to trigger it (or just speak if using always_active mode)

Activation modes

Agent name (recommended)

Set agentName in config. Say the name anywhere in a sentence:

• "Bobby, what's the weather?"
• "Can you help me bobby?"

Always active

Set activationMode to "always_active". Listens to everything. Best for solo use.

Wake word (prefix)

Set wakeWords (e.g. ["hey claw"]). Must be at the start:

• "Hey Claw, what's the weather?"

Commands

• /join — Bot joins your voice channel
• /leave — Bot leaves
• /mode wake_word or /mode always_active — Switch activation mode
• /status — Connection info, providers, active speakers, uptime

Config reference

Key	Default	Description
discordToken	—	Discord bot token (required)
sttProvider	"deepgram"	"deepgram", "whisper", or "whisper-local"
ttsProvider	"openai"	"openai" or "edge"
agentName	—	Name the agent responds to (e.g. "bobby")
agentId	"main"	OpenClaw agent to route voice messages to
activationMode	"wake_word"	"wake_word" or "always_active"
edgeTtsVoice	"en-US-AriaNeural"	Edge TTS voice
deepgramApiKey	—	Required if sttProvider = deepgram
deepgramLanguage	"en-US"	Deepgram language
openaiApiKey	—	Required if ttsProvider = openai
ttsVoice	"nova"	OpenAI TTS voice
whisperModelPath	—	Path to whisper.cpp model file
enableBargeIn	false	Interrupt bot when user speaks
wakeWords	["hey claw", "ok claw"]	Prefix trigger phrases

Troubleshooting

• Bot doesn't join voice: Make sure it has Connect + Speak permissions
• No transcription: For whisper-local, run whisper-cpp --help to verify it's installed
• No audio playback: Make sure edge-tts and ffmpeg are in your PATH
• Bot doesn't respond to name: Check agentName matches what you say (case insensitive)
• First response is slow: Normal — the voice agent needs a cold start. Subsequent responses are faster.
• Audio cuts off mid-sentence: Make sure enableBargeIn is false (default)