Voice Companion

Your agent reads and writes text. But the most natural interface for half of what you do is voice. Voice Companion bridges that gap.

What this does: Adds complete voice capabilities to any OpenClaw agent. Generate natural-sounding speech with ElevenLabs, transcribe incoming voice memos with Deepgram, and send voice messages across Telegram, WhatsApp, and other platforms.

Why voice matters for agents:

• Morning briefings you actually absorb. A 2-minute spoken summary while you make coffee beats a wall of text you skim and forget. Your agent reads your calendar, weather, and priorities aloud.
• Hands-free interaction. Send a voice memo, get a voice response. Useful while driving, cooking, or walking. The agent becomes a conversation partner, not a text box.
• Emotional range text cannot carry. A well-delivered story, a dramatic reading, a calm meditation guide — voice makes your agent feel present in a way text never will.
• Platform-native delivery. Outputs OGG Opus formatted specifically for Telegram and WhatsApp voice messages. Recipients see a proper voice waveform, not a file attachment.

Includes:

• Complete SKILL.md with setup instructions and all commands
• tts.sh — ElevenLabs text-to-speech wrapper with voice selection
• transcribe.sh — Deepgram Nova-2 speech-to-text (supports ogg, mp3, wav, m4a)
• voice-memo.sh — Full pipeline: generate speech, convert format, deliver to platform
• Voice selection guide with personality-matched recommendations

Requirements:

• ElevenLabs API key (free tier available, paid for premium voices)
• Deepgram API key (free tier includes 12,000 minutes)
• ffmpeg installed (for format conversion)

Use cases that work well: Storytelling and bedtime stories. Daily briefings. Podcast-style summaries of long articles. Meditation and breathwork guides. Voice responses in group chats. Accessibility for users who prefer audio.

Designed for OpenClaw. Works with any model.