A CodexBar alternative that answers your agents

CodexBar is a well-made free menu bar app: open source, actively maintained, showing usage windows, credits, and reset timers for a long list of AI providers. If the whole job is a limit meter in the menu bar, it does that job. AgentPeek starts from the same place, with live usage gauges in the menu bar: 5-hour and 7-day windows for Claude Code and Codex, Cursor's monthly window, and token totals for the agents that keep local records, read from what is already on your Mac with no extra sign-in.

The difference is everything past the meter. A usage number tells you an agent is running; it does not tell you which of your six sessions finished, which one has been sitting on a permission prompt for twenty minutes, or what any of them changed. AgentPeek monitors the sessions themselves:

• Live state per session: executing, thinking, waiting, or idle, with transcripts, diffs, and todos.
• Permission prompts and questions answered with a keystroke, without a window switch.
• Follow-up prompts sent into resumable sessions, with the reply streaming back.
• Notifications when an agent finishes, stalls, or crosses a budget.

AgentPeek also spends the Mac notch on your agents: a compact pill that expands the moment something needs you, with the prompt already laid out. The Agent Board floats a kanban of every session, widgets pin todos, transcripts, or usage to any display, and local dev servers get their own list. The menu bar stays available for Macs without a notch; it is one surface, not the whole app.

Scope. A limit meter and a full agent monitoring and response surface are different amounts of software, and the difference is maintained daily: session monitoring, prompt channels, and usage readers for twenty-five agents, updated as those agents change, with every release in the public log. The license is one-time, $15 at the limited-time price and regularly $19, with a free 3-day trial. The full cost math is on the build vs buy page.