Agent Speak
Speak messages aloud to notify the user when completing tasks.
Usage
agent-speak "Your message here"
agent-speak --worktree <path> "Your message here"
agent-speak --voice <name> "Your message here"
agent-speak --model <model-id> "Your message here"Voice Selection
Use --worktree with the current working directory to get a deterministic voice unique to your worktree. This ensures agents in different worktrees have different voices.
agent-speak --worktree "$(pwd)" "Task complete"Always use the worktree flag so your voice is consistent and distinct from other agents.
Available voice presets
If you need to specify manually:
rachel,bella,elli,freya,nicole,domi(female)josh,adam,sam,arnold,dave,fin,clyde(male)
agent-speak --voice josh "Build complete"When to Use
Speak notifications after:
- Completing a significant task
- Finishing a multi-step workflow
- Encountering an error that needs user attention
- Completing a build or deployment
Examples
# After completing a feature
agent-speak --worktree "$(pwd)" "Finished implementing the authentication flow"
# After a build
agent-speak --worktree "$(pwd)" "Build complete with no errors"
# After an error
agent-speak --worktree "$(pwd)" "Build failed. Check the terminal for details"
# After a refactor
agent-speak --worktree "$(pwd)" "Refactoring complete. Updated 12 files"Guidelines
- Always use
--worktree "$(pwd)"for automatic voice selection - Keep messages brief (under 20 words)
- State what happened, not technical details
- Don't speak for trivial operations (single file edits, running commands)
- Use natural phrasing, not robotic language
Setup
Users must configure their ElevenLabs API key:
agent-speak config set-key <elevenlabs-api-key>Or set the ELEVENLABS_API_KEY environment variable.
To debug configuration in an agent session:
agent-speak config show