--- title: Terminals with your AI: a window per task, in colour description: How your AI opens terminal windows: a window per operation, specialist or project, named and coloured, with one command on a Mac, Windows or Linux. url: https://instibrain.ai/works-with/terminal updated: 2026-10-27 site: https://instibrain.ai/llms.txt --- # A window per task, named and coloured. Say "open the website terminal", and a new window opens in iTerm2, Windows Terminal, GNOME Terminal, Konsole or xterm, coloured, with the task inside. What you can ask → Where the credential stays instibrain window · iterm2, windows terminal or a linux terminal · a window per target `team/tools/terminal/terminal.md` targets and colours asked · an example Open the SEO specialist terminal for the pricing page rewrite. the AI, from the page Opened window 7: SEO Specialist, pale blue, with your task as its first message. what instibrain heldThe target's row on the terminal page: emoji, path, colours. who uses it, and why ## Who opens windows by name. Anyone running more than one AI at a time. a developer ### Every task in a window you can tell apart. why Five tasks in five grey tabs look the same. asks Open the billing terminal and start on the retry bug. the head of ops ### A server checked from your desk. why The check needs Claude Code on the server itself. asks Open the production server's window and check the disk. an engineer ### A fact from code that lives on a server. why The answer is in a codebase only the server holds. asks Ask the api server which port the worker listens on. what you can ask ## Your AI opens a named window for each task, on your computer or a server. Always a new window. An open window is never renamed for a new task. Open a window by name An operation, specialist, project or the curator, each coloured. A window from the desktop button Claude or Codex, fresh or resumed, in the target's colours. A server's window SSH first, then Claude Code. Questions stop for you. Read a window before writing A blocking question would take the next newline as an answer. A fact from a codebase on a server That project's Claude Code answers over SSH, or says not found. A file to a server Sent by scp; the remote path comes back. where the credential stays ## No credential. One Mac permission, once. Windows open on your own computer. A server's login is yours. on your computer macOS asks once whether the launcher may control iTerm2. A new window is ready once Claude Code waits for input and its colour shows. in your brain Each space's terminal page keeps a row per target: name, path, colours. what instibrain never touches ## Your windows stay yours. Our server stores the terminal page, with names and colours. - Typing into a window you opened. The AI types only into windows it opened, by id. - A newline into a blocking question. Such a window is left to you. - A window found by what is in front. Each window is found by its own fixed id, set when it opens. the limits ## Your AI gives each prompt one of nine colours. Claude Code takes no other. | Limit | What it means | | Nine prompt colours | Red, blue, green, yellow, purple, orange, pink, cyan or default. | before you connect it ## Your AI types only into the windows it opened. **What does a terminal page have to do with credentials?** Nothing. It opens windows; macOS asks once for permission. **Can the AI type into my terminal?** Only into windows it opened, after reading the screen. **Which terminals?** iTerm2 on a Mac, Windows Terminal on Windows, and GNOME Terminal, Konsole or xterm on Linux. ## One window per task, never confused. Keep each target's row on the terminal page. Every window opens right. Request access → Connect Claude Code browsers and terminals Chrome Profiles and bookmarks, changed through Chrome's own API. every kind Works with Every system by kind, and the rule each page shares. --- instibrain is made by WebWork AI, Inc. Everyone uses the same models · the difference is what yours knows.