mdmxmyr/agent-ram-monitor

Agent RAM Monitor is a tool for people who run AI coding assistants like Claude Code, Codex, or Copilot CLI throughout the day and sometimes notice their computer's memory quietly filling up. Instead of just seeing a wall of unlabeled processes in Activity Monitor, this tool groups memory usage by coding session, showing you which specific task an agent was working on and how much memory its entire process tree, including any helper tools or browsers it spawned, is actually using. The tool comes as both a command line program and a native menu bar app for Mac. It breaks down each session's memory into categories like helper servers, browser windows, and test runner processes, and if a group of many identical worker processes is bloating things, it collapses them into a single line so you can kill the whole group at once. Each session is also linked back to its resumable ID, so once you kill a heavy process you can immediately restart that same conversation with the agent picking up right where it left off, now with a smaller memory footprint. Because memory problems often happen while nobody is watching, the app quietly logs a snapshot every few seconds to a small local history file. After noticing a spike, you can run a command that reconstructs what happened, showing exactly when memory jumped and which specific process caused it. The menu bar app also shows a live gauge of your Mac's total memory, turns a warning color when any single session crosses a memory threshold you can configure, and lets you jump straight to the window running that session. The project is written in Python using only its built in standard library, plus a single Swift file for the native Mac menu bar interface, so installing it does not require any extra Python packages. It only works on macOS, since it depends on Mac specific system tools, and it is released under the MIT license.