About LMM

Built for the Mac you already own

LMM shows which local AI models will run well on your Mac, with what settings and what to expect. When something slows down, it finds out why.

Check availability See the product tour

Why it exists

Everyone online has a huge rig. What about the Mac you own?

The Mac it started onApple silicon MacLocal AI, with the limits made clear

Watch a few local AI videos and the setup is usually the same: racks of GPUs, a maxed-out Mac Studio, a budget most of us don’t have. The model flies, and the video ends with “just run this.”

We wanted to know something smaller. What can I get running right now, on the laptop I already own? Or on a phone, or whatever else is on the desk? Which model is actually good, and which one only looks good in a chart?

Every answer lived somewhere different. One site for the model, another for the numbers, a forum thread for the settings, and a comment saying it crashed on theirs. Then the words started: quantization, context window, tokens per second. What does any of it mean? What do you gain by shrinking a model, and what do you lose?

Half the time it didn’t work. A model would download and not load. Or it loaded and crawled, with the Mac swapping to disk, and nothing said whether the model, the settings or the Mac was to blame.

LMM is the tool we wished we’d had at the start. It looks at your Mac, tells you what will actually run well, explains the settings in plain words and finds the problem when something is slow. It’s built for Macs, because that’s where we started.

What we had to learn

The words that stopped us

Local AI runs on a few ideas that rarely get explained first. Once you know them, the choices get a lot easier.

Quantization

Shrinks a model’s numbers so it takes less memory and runs faster. You gain room on your Mac. You lose a little accuracy, and the more you shrink it, the more you lose.

Context

How much text a model can keep in mind at once. A bigger context needs more memory, which is why a model that fits at 32k may not fit at 128k.

Tokens per second

How fast the answer is written. It’s only half of speed: the model also has to read your prompt first, and for a coding agent that can take a long time.

How we work

Honest answers, in plain words

Three rules we hold LMM to.

Honest answers over hype

If a setup will be Frustrating or Won’t fit, LMM says so before you press Load, and tells you what would help.

Plain English over jargon

When a technical word matters, LMM uses it and explains it, so the choices make sense.

Your Mac, your data

LMM runs on your Mac. Your prompts, conversations and files never come to us.

What LMM checks

Everything that decides how local AI feels

The Planner and Checkup look at the same things, on your actual Mac.

Which models and context sizes fit, and how well they’ll run
Memory, with room kept for macOS and your apps
Writing speed, reading speed and the wait for the first word
What’s slowing things down, from memory pressure to idle models
Ollama, LM Studio, llama.cpp and MLX servers
Headless Macs on your own network

Where it’s going

Local AI is an emerging field, and it’s moving fast

New models land every week, and the tools around them keep getting better at doing more with less. LMM’s job is to keep that progress clear: which new model is worth trying on your Mac, what settings suit it and how it performs once it’s running.

Better efficiencyStronger small models and smarter quantization fit more into the same memory.

Better toolsClear setup, clear settings and checks that explain themselves before anything changes.

Better harnessesAgents that use their context well get further on the Mac you already have.

Questions, ideas or a setup that won’t run?

Email us and tell us what you’re trying to do.

Email Us

[email protected]