Hey everyone 👋 I'm the maker. I run Claude Code and Codex side by side daily, and kept hitting the same two problems: an agent quits mid-task (network blip, or it just decides it's "done"), and sub-agents fudge their status when they hand work back. Polter is my fix — a terminal where one AI CLI supervises the others. It opens tabs, starts other CLIs, types into them, reads their screens. Workers report into a group chat + shared task list. If a worker stalls, the supervisor is told. If the supervisor goes quiet, it gets nudged too. Honest status: macOS + Claude Code is the path I use every day and trust. Codex/Qwen/opencode work as workers; supervisor mode for them is untested — if you try it, I'd love to hear how it goes. Windows does 63/72 core actions. Linux builds from source. Fully offline, no accounts, no API keys, MIT licensed. Happy to answer anything.

Polter

A terminal where one AI CLI supervises the other AI CLIs

A terminal where one AI CLI supervises the other AI CLIs