Baton

When you run more than one AI coding agent at once, the bottleneck shifts from the agents’ speed to your own attention. Baton addresses this by reading the state signals your macOS environment already emits — the docs describe this as ‘zero manual entry’ — and surfacing a sorted view in your menu bar: sessions where the agent is still running versus sessions where the baton is back in your hands. Click an item and you jump directly to that session. The install script handles dependencies; the core stack is a local Python app with a small web server backing the menu bar UI.

The differentiating detail is passive signal collection. Rather than requiring you to tag sessions, set timers, or file tickets, Baton reads what your machine already knows. The vendor describes it as reading ‘signals your machine already emits.’ That means onboarding is the install script — nothing more. For developers who resent tooling that demands its own upkeep, that is the entire pitch.

Baton fits a narrow but real scenario: a solo developer on macOS running two or more Claude Code or Codex threads who needs persistent, ambient visibility into which agent is blocked on them. It breaks down clearly outside that frame. macOS is a hard requirement — the docs describe no Linux or Windows path. The tool reads Claude Code and Codex state specifically; other agents or coding tools are not mentioned as supported. There is no multi-user view, no shared dashboard, and no API surface for piping status into another system. Teams doing any collaborative monitoring of agents need a different approach from the start.