LM Monitor

LM Monitor is a menu bar app for Macs that run local language models in LM Studio. While a model generates, the menu bar shows live tokens per second; while it reads a prompt, it shows prompt-processing progress; otherwise it shows the last request's speed. Click it for the full picture.

The popover shows a live card for each parallel slot, prefix-cache reuse, time to first token, context use with a truncation badge, and a model card with architecture, quantization, size, engine version and a memory-fit verdict: fits, tight or will swap. From there you can unload the model, start the server or check whether a model on disk will fit before you load it.

A System tab tracks CPU with the performance and efficiency core split, GPU utilization, GPU memory against the macOS budget, RAM, memory pressure, swap, disk, temperatures, and power for CPU, GPU, Neural Engine and DRAM, with 10-minute history graphs. History keeps your last 1,000 requests with per-model averages and CSV export.

Alerts catch GPU out-of-memory errors, failed model loads, a full KV cache, swap, thermal throttling, context overflow and runaway generations, and each alert tells you what to do. LM Monitor reads LM Studio's own logs, so it sees every request, whether it comes from the chat window, curl or another tool. It never reads or stores prompt or response text.