I created a simple terminal-based tool that uses a consumer Gemini, which is free, and turns it into a local AI agent that runs entirely on your system. The tool allows you to create, edit, and manage files, as well as code, and other system-related actions, such as opening applications, adjusting volume and screen brightness, and even system sleep, using plain language through a terminal interface.

It works 100% locally, with no backend and no need for a backend API key. It just leverages your existing Gemini web session. So nothing actually gets sent off to a server.

With all due respect, based on description, (seems like at least) you've have no idea what your bot has created.

Gemini CLI is open source btw: https://github.com/google-gemini/gemini-cli. (Codex too; Claude Code is closed.)