Terminal Basics

Basics · free

AITerm is a real PTY terminal. It's a normal zsh/bash/sh/fish shell first — you can use it exactly like any terminal — and the AI input bar is an optional layer on top. Nothing about the AI is in your way when you don't want it.

Tabs, panes, and workspaces

New tab — ⌘T. Each tab is its own shell session.
Next / previous tab — ⌘⇧] / ⌘⇧[; jump straight to a tab with ⌘1–⌘9.
Split the current pane — ⌘D (right) or ⌘⇧D (down). Panes can nest, so you can build a grid.
Move focus between panes — ⌘⌥ + an arrow key.
Close the focused pane (then tab, then window) — ⌘W.
Saved workspaces remember the whole pane layout and each pane's working directory, so you can reopen a project setup in one step.

Opening a saved workspace replaces the current layout. If a pane is running something or would be risky to replace, AITerm asks before closing it. You can also have AITerm restore your last session automatically on launch (Settings ▸ Terminal). Choose your shell and theme in Settings.

Clickable links

URLs in output are clickable — both the hyperlinks programs emit deliberately (the OSC 8 escape) and plain URLs AITerm spots in the text. Hold ⌘ and click.

Clicking never just opens something. In a terminal, the program printing the output chooses the visible text and the address behind it independently, so link text saying "release notes" can point anywhere. AITerm treats a click like any other risky action:

http / https — a confirmation shows the real host first, then the full address. Cancel is the default, so pressing Return does not open anything. If the address hides its destination behind an @, or uses a punycode host that can imitate a familiar name, the confirmation says so on its own line. An address whose host is written as a bare number — http://2130706433/, which a browser silently reads as 127.0.0.1 — is refused outright, because the confirmation could not honestly tell you where it goes.
file:// — revealed in Finder, never opened. Revealing cannot launch an app, run anything, or mount a volume, so a link to SomeApp.app shows you the file instead of starting it.
Everything else (ssh:, shortcuts:, mailto:, custom app schemes…) — refused, with the scheme named above the input bar. Those are the schemes where a click hands control to another program.

Opened, revealed, and refused links are all recorded in the audit log alongside command decisions — see Safety and Security.

The AI bar vs. typing directly

The input bar at the bottom is the AI-aware surface. When you type there, AITerm can classify the text, propose AI translations, show risk badges, and gate what runs.

You can also click inside the terminal pane and type on the shell line, like any terminal. In that mode the status shows the AI gate is off — you're driving the raw shell. Direct typing hands back to the AI bar when you click the bar, when a proposal is loaded, or after a short idle timeout. If you start typing directly while an agent is running, AITerm pauses the agent and resumes it when you return focus to the bar.

Input-bar sigils

AITerm classifies what you type before deciding to run it literally or translate it. You can force the decision:

A leading space forces a literal shell command (no AI translation).
A leading ? forces AI translation or question handling.
A leading : also forces AI translation.
A trailing ? is treated as a question for the AI.

If AITerm can't tell whether your input is a command or natural language, it runs it literally and can offer a translation afterward if the shell reports command-not-found — so a real command is never sent to the AI by surprise.