Your Mac, Your Model: The Complete Guide to Running Your AI Assistant on a Local Model (Series Recap)

Over the last eight posts, we've covered a power-user option in LaunchMy.ai: pointing your assistant at an AI model that runs on your own Mac instead of Claude in the cloud. We call it "bring your own AI model."

This final post pulls the series together. If you're new, start here. It lays out the honest requirements, how the pieces fit, the trade-offs, and which earlier post answers each question.

One thing up front. This is a power-user option. In the dashboard it's labelled "Advanced users only," and the label is accurate. Most people are better served by the standard setup, where your assistant uses Claude through your own Anthropic account. If you enjoy tinkering and care about keeping everyday conversations on your own machine, read on.

Series Recap: Start Here

Here are the eight earlier posts in a sensible reading order:

1. Bring Your Own LLM: Use Your Mac's AI Capabilities With Launch My AI covers what the feature is and why someone would want it.
2. Can Your Mac Run a Local AI Model? An Honest Hardware Guide explains memory, Apple silicon, and what "enough" really means.
3. How to Run Your AI Assistant on a Local Model on Your Mac: A Step-by-Step Setup Guide walks through the setup from start to finish.
4. Living With a Local AI Model on Your Mac: Practical Habits That Make It Work covers the day-to-day routines that prevent most problems.
5. Local Model or Claude? How to Tell If Your Mac's AI Is Earning Its Keep helps you decide when to switch back.
6. When Your Mac's Local AI Model Stops Answering: What Happens and How to Get It Back is the recovery playbook.
7. What Actually Stays on Your Mac When You Run a Local AI Model? A Plain-English Privacy Map maps out what stays local and what doesn't.
8. One Mac or Two Machines? Choosing the Right Home Layout for Your Local AI Model compares the two supported home setups.

If you only read two, make them the hardware guide (#2) and the recovery guide (#6). Together they answer "can I?" and "what happens when it goes wrong?"

The Honest Requirements

Before anything else, here's what you actually need. None of these are optional.

1. An Apple-silicon Mac with enough memory

The model runs on a Mac with Apple silicon. Intel Macs won't work, and neither will a Windows PC, a Linux machine or a Raspberry Pi.

Memory matters most. A large model needs about 15GB for itself, which fits comfortably on a 32GB Mac. A 16GB Mac can work, but only with a noticeably smaller model, and smaller models give weaker answers (more on that below). If your assistant will also run on that same Mac, it needs macOS 14 (Sonoma) or newer.

2. An assistant on your own hardware

This option is only available for assistants that run on hardware you own. You can set this up in one of two ways:

• On the same Mac as the model. This is the simplest setup and the one we recommend.
• On a Raspberry Pi or spare computer on the same home network as the Mac.

Only the model has to be on a Mac. The assistant itself can live on other hardware you own, as long as it's in the same home. Post #8 compares the two setups in detail. If you're new to running an assistant on your own hardware, our BYOD announcement explains how that works.

3. oMLX

oMLX is a free third-party Mac app for running AI models. It's the only model app we support. You install it, download a model and start it. Our dashboard reads the list of available models from oMLX, so you don't have to type any model names. You paste oMLX's key into the dashboard once.

4. Your Claude (Anthropic) key, still connected

This one is easy to miss. Even when your conversations run on your Mac, Dex, the built-in AI systems administrator that monitors and maintains your assistant, keeps using Claude through your own Anthropic account. That maintenance work costs much less than conversations do, but it isn't zero. You'll still need an Anthropic account and key.

What This Option Doesn't Do

A few clear limits save a lot of disappointment:

• It doesn't work with cloud assistants. Where your assistant lives is fixed when you create it. If you already have an assistant on a cloud server, you can't move it onto your Mac. You'd set up a new assistant on your own hardware.
• It doesn't work from outside your home network. Your assistant has to be on the same Mac as the model, or on the same home network as that Mac. Connecting to your home Mac over the internet isn't available.
• It doesn't fall back to Claude automatically. If your Mac sleeps, loses its network connection, or the model stops, your assistant stops answering. It stays silent until you fix the problem.
• We don't support your hardware or your chosen model. We check that your assistant can reach your Mac. How well a particular model performs on a particular Mac is up to you.

How the Pieces Fit Together

It helps to think of four parts:

1. Your assistant is what you chat with on Telegram, WhatsApp or email. It answers only contacts you've approved.
2. The model on your Mac does the thinking for everyday conversations.
3. Dex watches over the machine, sends your daily health report, checks for security patches, cleans up disk space and fixes problems you ask it about. Dex runs on Claude.
4. Your Anthropic key powers Dex's work, and powers your conversations too if you switch back to Claude.

Switching between your local model and Claude is a setting on your assistant's dashboard page. Moving from Claude to your local model keeps your existing conversation. Switching back to Claude is one click, but it starts a fresh conversation.

What It Costs

Choosing a local model doesn't change your LaunchMy.ai subscription. It isn't a separate plan or an add-on. The Starter plan is $5.99/month for your first 6 months with the launch offer, then renews at $9.99/month.

Here's what changes: everyday conversations no longer bill your Anthropic account per message. Their running cost is the electricity your Mac uses. You still pay for Dex's smaller Claude usage. Dex's cost monitoring and monthly budget alert help you keep an eye on that.

For the full picture of monthly costs across all setups, see How Much Does It Really Cost to Set Up Your Own Private AI Assistant Per Month?

The Real Trade-Offs

This is where a local model earns its "Advanced users only" label.

It's slower

In our testing, replies took about 10 seconds on average, ranging from 5 to 46 seconds. The first reply after the model has been idle or just started (a "cold start") can take much longer. If you're used to Claude's speed, expect a real difference.

Answer quality depends on the model

A local model is not the same as Claude, and smaller models fall further behind. In our testing, a less capable model once reported that a step was done when it wasn't. Weaker models also struggle with multi-step jobs, like working with files and pictures. Post #5 explains how to tell whether your model is up to the tasks you give it.

Your Mac must stay awake and running

This is the most important habit. If your Mac sleeps, restarts or drops off the network, your assistant stops answering. We usually detect the problem within about 12 minutes and send you one email.

On a one-Mac setup, there's a catch: that email can't go out while the Mac is asleep, because the assistant is asleep too. With the two-machine setup, the assistant sits on separate hardware, so a sleeping Mac doesn't also take the alert down with it. This is one of the main reasons post #8 exists.

oMLX doesn't reopen on its own after a restart

This is the most likely way things break. After your Mac restarts, for a software update for example, you have to reopen oMLX by hand. Until you do, your assistant can't answer. Post #4 covers habits that make this easier to remember, and post #6 walks through getting back up and running.

Is It Right for You? Three Quick Questions

1. Do you have, or plan to have, an assistant on your own hardware? If your only assistant is in the cloud, this option isn't available for it. You'd need to create a new self-hosted assistant.

2. Do you have an Apple-silicon Mac with 32GB of memory, or can you accept a smaller, weaker model on 16GB? If not, stay on Claude. The standard tiers in your dashboard are likely the better experience.

3. Are you comfortable with slower replies and occasional hands-on fixes? If you want an assistant that feels effortless, Claude is the better fit. If you value keeping everyday conversations on your own machine and don't mind reopening an app after a restart, a local model may suit you.

If you answered yes to all three, go to the step-by-step setup guide in post #3. We also have a public explainer page in the Power users section of our site.

Wrapping Up the Series

Bringing your own model fits the core idea behind LaunchMy.ai: you own the hardware, the accounts and the data. We handle the setup, and Dex handles the maintenance. A local model extends that ownership to the thinking itself, but it asks more of you in return: the right Mac, a little patience, and a habit of checking that oMLX is running.

For most people, the standard setup with Claude on their own Anthropic account is the right choice. For the curious minority with a capable Mac, it's a real option, and it's live today. Whichever you choose, you can switch back to Claude from your dashboard in one click, though switching back starts a fresh conversation.

Thanks for following the series. If you're still deciding, start with the overview.