Run coding agents in parallel

Opening a second Claude Code session takes one terminal tab. Knowing which of your six sessions finished, which one has been waiting on a permission for twenty minutes, and which one quietly stalled: that's the part that doesn't scale. AgentPeek tracks each supported session and puts the visible set in the Mac notch, labeled by project, with its state readable at a glance: executing, thinking, waiting, or idle, plus an attention flag when a session needs you. Advanced settings control how many session rows appear at once.

Views open one terminal window split into panes, each running the agent CLI you chose (claude, codex, cursor-agent, and the rest) in the folder you chose. Save a View for your standing setup, launch it, and every session appears in AgentPeek as supported activity arrives. Mixing vendors is the point: run Claude Code on the app repo while Codex works the API repo and Cursor handles a refactor.

When several agents want you at once, their permission prompts and questions stack in the notch; arrow keys move between them and a keystroke answers each one. The Agent Board keeps the wide view as a floating kanban: what's working, what needs attention, what's finished, sorted on its own while you work.

Parallel work burns quota in parallel. AgentPeek shows live usage gauges per agent: 5-hour and 7-day windows for Claude Code, the 7-day window for Codex, Cursor's monthly window, and token totals for the local-log agents, with a budget alert when reported usage crosses your chosen threshold (50 to 100 percent). Usage settings pick up to four agents for the expanded header and up to four for the collapsed pill, so the numbers you care about stay on screen.