The AI assistant

Describe what you want to do — “find the five largest files under /var/log” — and the assistant suggests a command with an explanation.

Asking for a command

Choose View > AI Assistant (⌥⌘A) or click the sparkles button. The panel appears below the terminal with the question field focused.
Type what you want to do and press ↩ (or click Suggest).
The suggestion appears as the command, an explanation, and — when the command is risky — an orange caution line.

Then use one of the buttons:

Insert (⌘↩) types the command into the terminal,
Copy puts it on the clipboard,
Save as Snippet… stores it in your snippet library.

Note Suggestions are never executed automatically. Insert leaves the cursor at the end of the typed command — read it, adjust it, and press ↩ when you are satisfied. Language models get flags subtly wrong; the last step is deliberately yours.

This is what the panel does on the on-device backend. On the Claude API backend it becomes agent mode, which runs commands for you — still showing each one before it runs.

Choosing a backend

In OliTerm > Settings…, the Command Assistant section offers two backends:

On-device (Apple Intelligence) — the default. Requests never leave your Mac and no account is needed. Requires Apple Intelligence to be enabled in System Settings.
Claude API — uses Anthropic’s Claude, and turns the panel into agent mode. Fill in the Claude API section: your API key (click Save Key; it’s stored in your login keychain, not in the app’s settings), the Model, and — only for identity-linked keys issued by an organization — the Workspace ID.

Choosing a model

The Model menu lists the models your key can actually use. It is filled from your account rather than from a list built into OliTerm, so a model released after this version still appears — click Refresh Model List to fetch the current set (it needs a saved API key). The list is remembered between launches, and a model you chose earlier stays selectable even if the account stops offering it, so nothing switches behind your back.

Sharing terminal context

Include recent terminal output in requests (off by default) sends the last few thousand characters of the current session along with your question, so you can ask things like “why did that fail?”. Leave it off if your terminal may show content you don’t want in a request — especially with the Claude backend, where requests go to Anthropic’s servers.

OliTerm Help