General local command execution
The Terminal capability is designed for ordinary command-line work rather than a fixed menu of domain actions. It can run diagnostics, scripts, builds, tests, pipelines, package managers, and other non-interactive zsh commands needed to answer the user's request.
Commands begin in the managed ~/.privateAI/workspaces/default workspace unless the user explicitly chooses another folder. Selecting a folder changes the authorized working root; terminal availability does not depend on selecting a document folder.
Visible supervision
- The App shows the command, working directory, elapsed time, checkpoint status, and Stop control.
- Standard output and error drain to bounded local logs instead of flooding model context.
- Long-running attached commands can return bounded checkpoints while output continues draining.
- Cancellation, timeout, App shutdown, worker failure, and transport timeout terminate the recorded process group.
- Tool calls are serial by default unless a concrete implementation explicitly opts into safe concurrency.
Current limits
The current terminal phase is non-interactive. It does not support persistent detached servers, PTY applications, privileged execution, or secret input. Passwords, tokens, private keys, and other secrets must not be included in prompts or Tool arguments.
Typical workflows
Examples include inspecting a project, running a test suite, checking a network route, invoking a compiler, processing local data, or diagnosing a process. Success is based on the real command result and final user-visible answer, not merely on the model selecting the Terminal Tool.