# AI use cases for Mlxtop (maximpri/mlxtop)

Repo: [maximpri/mlxtop](https://github.com/maximpri/mlxtop) | Use cases: 3 | Page: https://dedupedai.com/use-cases/maximpri-mlxtop/

mlxtop is a terminal dashboard that shows memory, paging, GPU and live tokens per second for local LLM runtimes such as oMLX, Ollama, llama.cpp and vLLM.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Find why a local model slowed down: https://dedupedai.com/use-cases/maximpri-mlxtop/find-why-local-model-slowed/
- Check a remote model server over SSH: https://dedupedai.com/use-cases/maximpri-mlxtop/check-model-server-over-ssh/
- Spot an Ollama model partly running on the CPU: https://dedupedai.com/use-cases/maximpri-mlxtop/spot-ollama-model-running-on-cpu/

## How to find why a local model slowed down with Mlxtop

Repo: [maximpri/mlxtop](https://github.com/maximpri/mlxtop) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README

Read memory, paging, GPU and runtime speed side by side to see which one changed.

**Who it is for:** A person running a local model on their own Mac or PC who notices replies have become slow and wants to know whether the machine or the model is at fault.

**The problem:** A slow local model can come from memory pressure, swapping, a full GPU or a throttled card, and the runtime's own logs rarely show this.

**The outcome:** A single verdict line on the Overview screen names the likely cause, backed by memory, paging, GPU and tokens per second readings.

### Steps

1. Install Mlxtop with `brew install maximpri/tap/mlxtop`, or use the install script from the README on Linux and macOS.
2. Start your runtime and send it a prompt so a model is loaded.
3. Run `mlxtop` to open the dashboard.
4. Watch view 1 · Overview while the model answers and read the verdict line.
5. Press `d` for diagnostics and setup help for the current runtime if a reading is missing.

### Prompt to paste into your agent

```text
Using the mlxtop repo at maximpri/mlxtop, run `mlxtop --once` on <target machine> while the local model served at <endpoint> is responding slowly. Report the verdict line and whether it names memory pressure, paging, a model partly on the CPU, a full card or heat throttling. Verify the report includes tokens per second and memory figures before you state a cause.
```

### Caveat

Mlxtop shows what changed when the model slowed down but cannot prove the cause, and GPU activity covers the whole system rather than one model.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/maximpri-mlxtop/find-why-local-model-slowed/

## How to check a remote model server over SSH with Mlxtop

Repo: [maximpri/mlxtop](https://github.com/maximpri/mlxtop) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Print a text report of a headless machine's local LLM health without opening a dashboard.

**Who it is for:** An operator or an AI agent managing a headless Mac, Linux box or GPU server that runs a local model and is reachable only over SSH.

**The problem:** Opening a full dashboard on a server is awkward, and automated checks need plain text output they can read and compare.

**The outcome:** A text report from `mlxtop --once` shows the runtime readings, and `mlxtop doctor` confirms which readings are available on that machine.

### Steps

1. Install Mlxtop on the server with `brew install maximpri/tap/mlxtop` or the install script for that platform.
2. Connect over SSH to the server.
3. Run `mlxtop doctor` to check the runtime connection and which readings are available.
4. Run `mlxtop --once` to print a text report and exit.

### Prompt to paste into your agent

```text
Using the mlxtop repo at maximpri/mlxtop, connect to <target machine> over SSH, run `mlxtop doctor` and then `mlxtop --once`, and return both outputs. Verify that the doctor output lists the runtime connection status before you summarise memory and tokens per second from the report.
```

### Caveat

Mlxtop monitors one server at a time, and polling is sampled, so a request that starts and finishes between two samples can be missed.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/maximpri-mlxtop/check-model-server-over-ssh/

## How to spot an Ollama model partly running on the CPU with Mlxtop

Repo: [maximpri/mlxtop](https://github.com/maximpri/mlxtop) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

See which loaded Ollama models use VRAM, context size and CPU or GPU placement.

**Who it is for:** A developer on a PC with an NVIDIA or AMD card who runs Ollama and suspects a model is spilling onto the CPU.

**The problem:** When part of a model sits on the CPU, generation becomes much slower, and Ollama does not make that placement obvious at a glance.

**The outcome:** The Overview verdict names a model partly on the CPU, and the loaded model list shows its VRAM and CPU or GPU placement.

### Steps

1. Install Mlxtop with the install script for Linux, or `brew install maximpri/tap/mlxtop` on macOS or Linux.
2. Start Ollama and load a model by sending it a prompt.
3. Run `mlxtop` and open view 1 · Overview.
4. Read the loaded models and their CPU/GPU placement, and check the verdict line for a model partly on the CPU.
5. Use `[` and `]` to select a GPU card if the machine has more than one.

### Prompt to paste into your agent

```text
Using the mlxtop repo at maximpri/mlxtop, start a loaded Ollama model on <target machine>, run `mlxtop --once`, and report the loaded model name, its VRAM use and its CPU/GPU placement. Verify the report shows the placement before you say whether the model is partly on the CPU.
```

### Caveat

For Ollama, speed is recorded after each request completes, so live tokens per second are not shown while a reply is still streaming.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/maximpri-mlxtop/spot-ollama-model-running-on-cpu/