What is cicero?

Cicero is a self-hosted voice interface you install alongside a coding agent you already use, such as Claude Code, Codex, or Gemini. You speak to it, it answers out loud, and the agent underneath does the actual work of writing code, fixing bugs, and opening pull requests. Say something like "fix the failing auth test and open a PR" and Cicero acknowledges within about a second, the coding work happens in the background, and it tells you out loud once the pull request is ready. Commands you have marked as needing approval, like pushing to git, wait for a spoken yes before running. There are two ways to use it. The lightweight sidecar mode attaches to a coding agent session you already have running and simply speaks its replies out loud, with no models to download and almost no setup. The full setup adds a real spoken conversation you can have from any browser on your local network, plus an optional Telegram add-on that lets you talk to your agent over an actual phone call. In the full setup, speech recognition and speech synthesis run locally on your own machine, so audio does not need to leave your computer. Cicero can clone a voice from a single sample recording and speak in that voice going forward. You can talk over it mid-sentence and it will stop and listen, using a small local model to confirm a real voice interrupted it rather than background noise. It can also read the emotional tone of what you say and pass that along to the agent. A more advanced setup lets you run several agents behind one call, each with its own voice, so you can say things like "let me talk to the coder" to switch between them. A GPU is recommended for the fastest response times but is not required. Cicero also works on plain CPU, on Apple Silicon Macs, and on Windows machines with an NVIDIA GPU. The project is under active development, with the browser, phone, and sidecar modes already working.