agentbuzz

AI coding agents run for minutes at a time. When one stops to ask permission, it freezes — and you usually find out ten minutes later when you happen to look. agentbuzz puts that moment on your iPhone and Apple Watch. APPROVE FROM YOUR WRIST When your agent asks to run a command, the notification carries the buttons. Tap Allow and the run continues. Tap Reply and dictate the next instruction instead. Your laptop stays shut. KNOW WHEN IT FINISHES OR BREAKS A run that finishes gets a one-line summary of what actually changed and how long it took — and on Claude Code, the files touched and tools used as well. A run that fails says so. QUIET BY DEFAULT A naive notifier fires every time the agent stops talking, which is a buzz every twenty seconds. agentbuzz only speaks when a run crosses your threshold — 90 seconds by default. Short turns stay silent on purpose, and the app tells you when it deliberately stayed quiet, so silence never looks like something broken. YOUR CODE NEVER LEAVES YOUR MACHINE The summary is built locally, on your computer, from a transcript that never moves. Only that one line is sent — no file contents, no diffs, no prompts, no tool output. The local script is open source so you can read exactly what it reads and exactly what it sends. SAFE BY DESIGN Commands that can destroy work or leak credentials cannot be approved from a notification at all. They open the app and show the full command text first. If a decision does not reach us in time, nothing is denied — your terminal simply asks you as it normally would. SETUP Requires a computer running Claude Code, Codex CLI, or Cursor. Run one command in your terminal and it wires up whichever of them it finds. Enter the code it prints and you are done. No account, no credit card. Permission prompts work on Claude Code and Codex CLI. Cursor exposes no hook for them, so there it reports finished and failed runs. Free while it is new.