Open-source tool lets AI coding agents talk back to you by voice
Cicero is an open-source project that enables two-way voice conversations with AI agents such as Claude Code, Codex, Gemini CLI, or any tool that speaks ACP (), as well as . Unlike existing one-directional options like Claude Code's voice feature or dictation tools such as Wispr Flow, which only capture what a person says, Cicero lets the agent initiate contact — speaking up when a task finishes or even placing an actual phone call. It works especially well with Hermes Agent and supports multiple profiles, each with its own TTS voice and personality.
Users can receive a Telegram bot message or a call, and can interact either locally or remotely through a browser. Running (STT) and (TTS) locally keeps audio from ever leaving the machine. The developer reports achieving roughly a 1-second response time by using streaming.
The same developer also maintains tts-bench and a blind-voting TTS arena for comparing open models.
Key points
- Cicero enables two-way voice conversations with ACP-compatible agents (Claude Code, Codex, Gemini CLI) or
- Works especially well with Hermes Agent, with per-profile voice and personality settings
- Can send a Telegram bot message or place an actual phone call
- Running and synthesis locally keeps audio on-device
- Achieves roughly 1-second response time via streaming