Run Claude Code and Codex from one window

Choose the name before the first prompt, then the computer, Claude Code, Codex or Grok, and the account. A month from now you will still know what “billing refactor” was.

Pick up any session where it stopped, including ones you started in VS Code.

Paste screenshots with Ctrl+V, attach files or drop them in. The model and reasoning effort come from the list the CLI itself offers, so the choices are never out of date. Your draft is saved as you type.

Write the next instruction while the agent is still busy. It goes out when the turn ends, even if you have closed the window. Waiting prompts can be edited, reordered or removed, and “send now” pushes one into the running turn. Works with Claude Code and Codex.

When Claude Code asks you something, the question shows up as a proper form: single choice, multiple choice or free text. A kind of question we have not seen before still gets a reply box. Codex, in the mode AigoPilot runs it, does not ask questions.

“Stop after this step” waits for a safe point. “Stop now” interrupts at once, and a second button closes the session completely.

Switch any of these without restarting the session. The change is shown only after the CLI confirms it, so the screen never names a model you are not actually using. Plan mode is a toggle in the prompt box. For Codex, a new model applies from the next turn.

Type “/” and you get the CLI's own commands for that project: built-in ones, plugin commands, and your project and user commands. /compact and /clear do what they do in the terminal.

One click and the session writes a handover note: the goal, what is done, the decisions and why they were made, what is still open, the files that matter and the next step. A new linked session opens, reads it and waits for you.

The button shows what the handover will cost before you press it, and if anything goes wrong the old session is left as it was. In one test a session of about 150,000 tokens handed over with all 11 of its decisions intact. That test was with Claude Code. Codex sessions have the same button since October 2026, and there it has so far run only against a test program.

A session with a full context gets a card with three ways out: compact it, hand it over, or move to a model with a bigger window. And if a session has been idle longer than its prompt cache lasts, you are warned before you send, with the number of tokens that will be read again.

A session that hit its usage limit continues by itself one minute after the reset, even if its program was closed or the PC restarted in between. It is on by default. Turn it off everywhere or per project, or keep it for high-priority projects only. AigoPilot or its background engine has to be running at that moment. Tested with a test program, and on real limited sessions read from their transcripts. We have no record yet of a real usage limit being continued by AigoPilot.

Give each project its own account and new sessions use it. AigoPilot never switches accounts on its own. Signing in happens in Claude Code's own window. We have not yet seen a second account signed in and used for a session here; only the usual account has been.

The same per-project choice of account, for Codex and for Grok. It is built, with each account in a folder of its own, but we have not yet seen a second account signed in and used for a session. Until then it stays marked in development.

New Claude Code sessions reply in your language and in its own alphabet. Type Persian in Latin letters and the answer comes back in Persian script. The aim is text that reads as if it was written in that language, rather than translated into it. Code, comments and commit messages stay in English unless you change the setting. Claude Code only, with 21 languages to choose from.

Start, continue and watch Codex sessions as they stream, with the model list Codex offers on each computer and its reasoning summaries shown. Codex asks the studio before every command and every change to a file, through Codex's own protocol, and AigoPilot answers by the permission mode you picked, as it does for Claude Code. This has been seen on the real Codex CLI. The question forms work with Claude Code only.

Prefer the real terminal? A setting starts the unmodified Claude Code in its own console window, and AigoPilot follows the session from its files. So far this has been tried with a test program standing in for Claude Code, not yet with the real one started from the studio.

A session that is open somewhere else opens read-only first. You are asked before AigoPilot takes it over.

Grok is the third agent in the studio. AigoPilot reads the sessions that xAI's Grok CLI keeps on disk, so they sit in the same projects and lists as the rest, and it starts and continues Grok sessions through that CLI with your own xAI account. The model and the reasoning effort come from Grok's own list, and its tokens go into the same ledger, with the value Grok itself reports for each turn.

Every tool call of a Grok session is put to the studio first and answered by the permission mode you picked, as for Claude Code. Seen with the real Grok CLI: in the fully safe mode a session asked before it wrote a file and before it ran it, and its second turn remembered the first. Grok has no plan mode in the studio.

Grok can be installed and signed in from the studio with xAI's own installer, and its sessions go with a project you move to another computer. Both are built and tested, and have not yet been used on a real second computer. AigoPilot is built so that more agents can be added. None besides these three is built.