The author introduces Cicero, an open-source project that provides self-hosted, bidirectional voice interaction for AI agents. Unlike one-way dictation tools, Cicero allows agents to speak back and call users via Telegram, keeping all audio local.

  • It integrates with Hermes Agent and any system supporting the Agent Client Protocol (ACP), including Claude Code, Codex, and Gemini CLI.
  • The project supports multiple profiles, allowing each agent instance to have its own TTS voice and personality.
  • Implementation details reveal that streaming replies sentence-by-sentence achieves ~1s response times, while trimming silence from reference clips improves voice clone quality more than changing models.
  • Effective barge-in functionality requires a speech detection model rather than simple energy-based voice activity detection to avoid false triggers from typing.

Cicero enables users to maintain privacy by running speech-to-text and text-to-speech locally while interacting with agents through audio calls or Telegram bots.