Terminal Mode

Terminal mode runs the selected provider CLI in an interactive pseudo-terminal (PTY). Choose it in the card's Execution section when the native CLI experience matters more than Stib's structured agent stream.

When to use it

• The provider exposes an important interactive flow only in its terminal UI.
• You want to operate the CLI directly while keeping the card, repository, and conversation context together.
• The run should use the provider's native subscription/session behavior.

Use Agent mode when you need structured messages, portable questions, detailed tool rendering, reliable context telemetry, or automation based on agent events.

Behavior

The server starts the PTY in the card's repository or worktree. Input and resize events travel over an authenticated WebSocket. The terminal remains associated with the card and can be cancelled from the same execution controls.

Closing or refreshing the client is not equivalent to terminating the server-side process. Reopen the card to inspect its current state.

Limits

• Arbitrary terminal output cannot be parsed as reliably as an Agent-mode stream.
• Some provider actions, tool calls, questions, and context values may appear only as terminal text.
• Interactive full-screen applications depend on provider and terminal compatibility.
• Automation should rely on explicit process/session completion, not a phrase printed by the CLI.

Review the Git diff after every terminal run, especially after cancellation or network loss.

Next: Agents and Git Integration.