Launch offer: Rollcage for $29 $39. Ends on 18 October. 36 spots left at $29. Get it →
Rollcage

Know when your AI agent needs you, and how much of your plan is left

Agents stop to ask permission while you're in another window, and long runs finish without telling anyone. Rollcage follows each session through the agent's own hooks, tells you the moment one needs you, and shows how much of your plan's 5-hour and weekly limits is left.

Needs you, working, done

The menu bar lists every running agent and what it's doing. A dot on the Rollcage icon means one is waiting for you.

  • Needs you: a permission prompt or a question. The notification says what the agent wants to do.
  • Done: the agent finished its turn. Only turns longer than 30 seconds are announced, so the quick replies you're watching stay quiet. You choose the number.
  • Click a notification, or the arrow next to the agent, to bring its terminal or editor to the front.
AgentHooks Rollcage uses
Claude CodeUserPromptSubmit, PostToolUse, Notification, Stop
Codex CLIUserPromptSubmit, PostToolUse, PermissionRequest, Stop
CursorbeforeSubmitPrompt, afterShellExecution, afterFileEdit, stop
Gemini CLIBeforeAgent, AfterTool, Notification, AfterAgent

They're installed together with the Guard and only record what the session is doing. They never change what the agent does. Codex asks you to approve new hooks once with /hooks.

Plan limits

Claude Code passes your Pro or Max plan's 5-hour and weekly usage to its status line. Turn on Track Claude limits in Rollcage › Usage and Rollcage reads the numbers there, then runs your own status line exactly as before. If you didn't have one, Claude Code shows a short line with your usage. Codex limits are read from its session logs, with nothing to set up.

  • Bars for the 5-hour and weekly windows, with when each resets.
  • A projection: at your current pace, when you'll reach the limit, if that's before the reset.
  • Notifications at 80% and 95%.
  • If you like, the 5-hour percentage next to the menu-bar icon.

The numbers are the ones Claude Code and Codex last reported on this Mac. Claude reports them on Pro and Max plans after the first reply in a session, so usage on claude.ai or another computer shows up with the next reply here.

Questions

Does Rollcage need my Claude login or API key for the limits?

No. It reads what Claude Code already passes to its status line and what Codex writes to its session logs. Nothing is sent anywhere.

Will it replace my own status line?

No. Rollcage runs your status line and shows its output unchanged. Turning plan limits off puts your original setting back.

Can I turn off the alerts for finished turns?

Yes, separately from the alerts for agents that need you, in Rollcage › Settings › Agent alerts.

Keep reading