Claude Codesessions in parallel is a productivity win until you realize one of them has been idling on a permission prompt for ten minutes while you were focused on another window. The mental overhead of constantly cmd-tabbing just to check if your LLM agent is actually working or stuck is surprisingly draining. Since I couldn't find a software solution for this, I decided to build a visual indicator that mimics hardware LEDs right on the desktop.
This tool essentially acts as a status light for your AI workflows. Instead of hunting through terminal tabs to see if a process is still running or waiting for input, you get a subtle visual cue beside the macOS Dock. It's inspired by a hardware gadget called SidePulse, but implemented as a lightweight software layer for those of us who prefer not to clutter our desks with more cables.
For anyone looking to integrate this into their AI workflow, here is the high-level logic of how the status tracking works:
How the status monitoring works #
The app doesn't just "guess" the state; it needs to hook into the process output or a specific API to determine the agent's current activity.
-
Process Polling: The app monitors the active terminal sessions or specific log files associated with the agent.
-
State Mapping: It categorizes the agent's state into three primary visual modes:
-
Green/Pulse: The agent is actively processing or generating code.
-
Yellow/Static: The agent is idling or waiting for a network response.
-
Red/Flash: The agent has encountered an error or is awaiting a user permission prompt.
- UI Rendering: A transparent overlay is positioned precisely next to the macOS Dock, rendering a series of "LED" pixels that change color based on the mapped state.
If you are building your own custom LLM agent or using a tool like Claude Code, having this kind of peripheral awareness is a massive quality-of-life improvement. It transforms the experience from "I hope this is still running" to "I can see it's still thinking" without breaking your flow state. Since this is a niche utility, the deployment is straightforward. You basically run the monitor in the background, and it keeps the "LED strip" updated in real-time. It's a simple piece of kit, but it solves a very specific pain point for anyone doing heavy agentic coding. It turns the invisible state of a background process into a concrete visual signal, which is exactly what's needed when you're managing complex, multi-step AI tasks.
Apple is bringing Alibaba's Qwen AI to Mac users in China 2d ago
Typing is actually a thinking process that voice input can't 2d ago
Coding is no longer about syntax when LLMs can generate a 2d ago
Stop trusting AI to write your final commit because LLMs are 2d ago
Claude Code is switching to Auto Mode by default because humans 2d ago
Grindr claims AI is now doing the work of 200 engineers 3d ago
Next Nvidia and Wall Street are teaming up for a $500B AI financing →