---
title: Use terminals
description: Run commands in the terminal drawer, scoped to the open session.
section: Guides
order: 4
---

# Use terminals

The terminal drawer runs at the bottom of the window. It gives you a real shell on the host device, next to the transcript.

## Open and use the drawer

1. Open the terminal drawer from the bottom of the window. Drag its top edge to set the height.
2. Open a terminal tab. Each tab runs a shell in the session's project folder on the host device.
3. Run commands as usual. Output streams live.

Terminals belong to the open session. Switch sessions, and each one keeps its own tabs.

## Tab behavior

- **Reorder** tabs by dragging them.
- **Close** a tab with a middle click.
- **Replay.** When you reopen a terminal or reconnect, the drawer replays the recent output, then continues live.
- **Limits.** Up to 32 terminal tabs per session. A tab whose shell exited stays for 30 minutes, then clears.

## Why a terminal next to the agent

The agent runs its own commands inside the transcript. The drawer is for you: start a dev server the agent should not own, run a command you do not want in the session history, or poke at state while you read a diff.

## Next steps

- [Start a session](start-a-session.md): give the agent work worth supervising.
- [Review changes](review-changes.md): read the results.