ModelPeek See your local LLM runners and the models they have loaded.
Ollama, LM Studio, MLX and llama.cpp: which models are loaded, and the memory, CPU and GPU they use.
- Status
- Available now
- Price
- Free · MIT license
- Requires
- macOS 13 or later
- Permissions
- None
- Network
- This Mac only
- Account
- Not needed
- Telemetry
- None
The problem
Your Mac has 9 GB less free memory than it should. Is an Ollama model still loaded from an hour ago, is LM Studio holding one, or is an MLX server you forgot about still running? Each runner has its own command or window, and none of them shows what it costs the rest of your Mac.
The solution
ModelPeek finds the local LLM runners that are running and shows one card for each, with the models it has loaded and the memory, CPU and GPU its processes use. Ollama and LM Studio are asked through their own APIs on 127.0.0.1, which give the model, its size, how much of it is on the GPU and when it will unload. MLX, llama.cpp and llamafile are read from the model they were started with. Nothing about the rest of their command lines is kept or shown.
Features
- Ollama, LM Studio, MLX (mlx_lm), llama.cpp and llamafile, recognised by their processes
- Loaded models with size, GPU/CPU split, quantization and context length where the runner reports them
- Memory footprint, CPU and GPU share of each runner, from macOS's own counters
- The runner's processes with their roles (server, model runner, app)
- Local only the runners' APIs are asked on 127.0.0.1 and nothing leaves your Mac
- Command lines are never stored or shown, only the model name
How it works
- Open ModelPeek from the launcher.
- See which runners are running and which models they have loaded.
- Compare their memory and GPU use, then unload a model in its runner if you need the memory back.
Privacy
- Reads
- The process list (ps), only while ModelPeek is open; of a runner's command line only the model argument is used, never shown or stored
- Memory and CPU time of runner processes your user can inspect (libproc), and their GPU time (ioreg)
- Ollama's and LM Studio's local APIs on 127.0.0.1 (ports 11434 and 1234) while their processes run
- Access
- None.
- Network
- Everything stays on your Mac: no account, no telemetry, and nothing is sent to any server. ModelPeek asks Ollama and LM Studio which models they have loaded through their own APIs on 127.0.0.1, which never leaves your Mac.
Installation
ModelPeek is part of MacPeek. Install MacPeek, open Manage utilities and make sure ModelPeek is switched on (it is on by default). There is nothing separate to download.
Open source
MacPeek and every utility in it are MIT-licensed. Read the code on GitHub, or see how a utility is built.
Frequently asked questions
What does ModelPeek do?
ModelPeek finds the local LLM runners that are running and shows one card for each, with the models it has loaded and the memory, CPU and GPU its processes use. Ollama and LM Studio are asked through their own APIs on 127.0.0.1, which give the model, its size, how much of it is on the GPU and when it will unload. MLX, llama.cpp and llamafile are read from the model they were started with. Nothing about the rest of their command lines is kept or shown.
What does ModelPeek read from my Mac?
The process list (ps), only while ModelPeek is open; of a runner's command line only the model argument is used, never shown or stored; Memory and CPU time of runner processes your user can inspect (libproc), and their GPU time (ioreg); Ollama's and LM Studio's local APIs on 127.0.0.1 (ports 11434 and 1234) while their processes run.
Does ModelPeek need any permissions?
None.
Does ModelPeek send my data anywhere?
No. Everything stays on your Mac: no account, no telemetry, nothing is sent anywhere. ModelPeek asks Ollama and LM Studio which models they have loaded through their own APIs on 127.0.0.1, which never leaves your Mac.
Is ModelPeek free?
Yes. ModelPeek is part of MacPeek, which is free and open source under the MIT license.
Get ModelPeek with MacPeek.
Free and open source.