minjun1177/aetheris ? reverse-engineered prompt

Build me a Python terminal AI assistant that can work in the current folder, read and edit files, run shell commands, search the web, and answer back in the terminal.

I want it to work well with small local models first, especially Ollama, but also let me connect to hosted providers like Anthropic, OpenAI, Gemini, and Attacca with the same commands and approval flow. It should ask before doing anything risky, keep track of permissions, and be able to undo the last file change it made. If possible, make it save sessions so I can resume them later, and include a few simple slash commands like connect, undo, clear, sessions, and settings changes.

It would be great if every edit can be tied to git so rollback is easy, and if the tool can verify changes after it runs. Look up current docs online if you need to.