Integrations
Command Assistant
Command Assistant turns a description into an editable shell command. Available in Terminalis 3.4.0 and later, it uses your connected ChatGPT account and the active terminal's shell and current folder. You review the result at the prompt and press Enter separately to run it.
Connect ChatGPT
- Open Settings with the sidebar gear or Cmd+,, then select AI → Generate commands.
- Choose Add ChatGPT connection. You can give the connection a name to recognize it later.
- Choose Connect in browser and complete the browser sign-in.
- Return to Settings and wait for Connected and for model loading to finish. Check the Model selection and choose an available model if needed.
- Keep Default connection selected for the connection you want to use normally.
ChatGPT is the supported provider. Sign-in, model discovery, and command generation need an internet connection. The distributed app includes its helper; no separate Node.js or Codex installation is needed. Ordinary terminal use does not require an AI connection.
Generate a command
- Select a terminal and start at an empty, idle shell prompt. Bash, Zsh, and Fish are supported; their shell integration must have finished initializing.
- Press Cmd+I or choose Action → Generate command….
- Describe the command, for example: “Find files larger than 100 MB under the current folder.”
- Press Enter to generate. The footer shows the selected connection and model.
- When generation succeeds, the assistant closes and inserts the command at that terminal's prompt. Read it and make any edits, then press Enter again when you want to execute it.
The result is a single-line shell command. Generation does not run it or add it to a saved-command library. To keep a useful result, copy it into a terminal command or a global command.
Use the assistant from the main terminal window. A running program or full-screen terminal app must return control to the shell first. Plain Ctrl+I continues to belong to the shell.
Clarification and cancellation
If the assistant needs a path or another essential detail, it asks a short question. Type your answer in the same field and press Enter; the next request includes your original description and the clarification exchange.
Press Escape, choose Cancel while generating, or use the close button to dismiss the assistant. Clicking back into the terminal also cancels it. A cancelled request cannot insert a late result. Errors keep your request available to edit and retry.
Each terminal keeps its own draft and pending request when you switch tabs. Returning to the original terminal restores its assistant state; a result belongs to that terminal even if you select another one while waiting.
Terminal changes
Terminalis checks that the originating shell and prompt still match before inserting a result. Typing into the shell, changing directory, running a program, or replacing the shell can prevent insertion. This keeps a command generated for one terminal state from arriving in a different one.
If a completed command cannot be inserted because the terminal changed, the assistant displays Terminal changed while generating and offers:
- Copy to copy the result for manual review and pasting.
- Regenerate to repeat the request using the terminal's current context.
- Dismiss to close the assistant.
Return to an idle prompt before regenerating. If delivery could not be confirmed, inspect the terminal before retrying so you do not insert the same text twice. See Command Assistant troubleshooting for connection, model, and prompt errors.
Manage connections
In Settings → AI → Generate commands, you can rename connections, select a model, and change the default connection. When more than one connection exists, use the assistant's AI connection picker to choose one for that terminal. Each connection uses its model selection from Settings.
Opening AI settings reloads the connected account's model list. Refresh models reloads it on demand. If a saved model is no longer available, choose another after loading finishes. Reconnect renews a sign-in that needs attention.
Disconnect removes the locally stored sign-in credentials for that connection while keeping its configuration. Remove disconnects it and removes its entry. Resetting general preferences preserves AI connections; manage them in the AI tab.
Data sent to the provider
A generation request sends your description, any clarification exchange, the shell name, operating system, and current folder path to the selected provider. Terminalis does not automatically include terminal output, scrollback, shell history, environment values, directory listings, or file contents. Anything you type into the request or an answer is part of the text sent to the provider.
Connection and model choices are saved in local settings. Sign-in credentials are stored separately under the app's configuration directory and are excluded from portable .terminalis exports. The provider handles submitted data under its own policies. See the Privacy Policy for the app and website's data practices.