Running AI on Your Own Computer: What to Prepare First
A beginner's guide to local AI with Ollama and LM Studio, covering installation requirements, how to check your memory, and what still works without an internet connection.
Running AI on your own computer takes three things: a program to run the model, the model file itself, and enough memory and storage to handle both. Install Ollama or LM Studio, download a model, and you can start chatting right away. With LM Studio, if you download your models in advance, core features like chat and document conversations keep working even without an internet connection.
What does “running it on my computer” actually mean?
Local AI means running a model you have downloaded directly on your own machine. Think of it like saving music files to a device and playing them there. Installing the program alone does not mean you are ready, though. You still need to download the model file you plan to use.
It helps to separate the name of the program from where the model actually runs. Ollama offers cloud models alongside the ones that run on your computer. So the fact that you are using Ollama does not by itself mean the AI is running locally. When this article talks about running on your computer, it refers to cases where you have chosen a local model.
The first thing to try is simply talking with a model. LM Studio also lets you attach a document and have a conversation based on its contents. There are features for connecting AI to other programs as well, but for beginners, chat is a fine place to start.
Can I install it on my computer?
As of September 2026, both tools support macOS, Windows, and Linux. The details differ, though. On Mac in particular, you need to check the chip type along with the OS version.
| OS | Ollama | LM Studio |
|---|---|---|
| macOS | 14 or later, supports Apple M series. Intel Macs run on CPU | 14 or later, supports Apple Silicon. Intel Macs not supported |
| Windows | Windows 10 22H2 or later, Home or Pro | x64 or ARM supported. x64 CPUs require AVX2 support |
| Linux | Packages documented for x86-64 and ARM64 | x64 and ARM64 supported, requires Ubuntu 20.04 or later |
If you use an Intel Mac, this difference decides which tool you can pick. LM Studio does not support it at all, while Ollama can run on the CPU. If you go with LM Studio on Windows, check whether your CPU supports the AVX2 instruction set.
Storage deserves a separate look. On Windows, Ollama needs at least 4GB of storage just to install the program. That figure does not include models. You need to set aside additional space for the model files you download.
How much memory do I need?
It is hard to tell whether you can run every model just by looking at your memory capacity. The actual requirement varies with the model file and your runtime settings. It also helps to read a program’s recommended specs and a specific model’s memory guidance as two separate things.
As of September 2026, LM Studio recommends 16GB of memory or more on Mac. If you have 8GB, it advises choosing a small model and setting a shorter limit on how much text can be read at once. On Windows x64, it recommends 16GB or more of memory and 4GB or more of dedicated graphics memory. Recommended specs are not hard minimums you must meet.
You can see how size changes things in Ollama’s guidance for Llama 2. It offers rough benchmarks of at least 8GB of memory for the 7B model, 16GB for 13B, and 64GB for 70B. Here, “B” is a marker used to distinguish model size. These numbers are specific to Llama 2, so you cannot apply them directly to other models.
There are also ways to estimate the memory you will need before you run anything. LM Studio estimates memory usage before loading a model. Rather than deciding based on a model’s name or size label alone, it is better to check against the actual model and settings you plan to use.
What order should I try things in?
If this is your first time, LM Studio is a convenient option, since you can search for models, download them, and chat all within the app. What follows is not a walkthrough of menu locations but the order of preparation and use.
- Check your computer’s specs. Look at your OS version, chip type, memory, and remaining storage.
- Install the program. Use the support requirements above to pick the tool that fits your machine.
- Download a model. In LM Studio, you can search for and download models inside the app. If you are short on memory, start with smaller models.
- Start with chat. Talk with the model first, then attach a document and try a conversation around it if you need to.
Ollama also lets you download models and use them conversationally. If you want to use it with other programs, you can connect them to each other inside your computer. Both tools have this kind of connection feature, so you can look into it once you are comfortable with chat.
How far can I get without internet?
With LM Studio, if you download your model files in advance, core features like chat, document conversations, and program connections work without an internet connection. If you plan to use it somewhere offline, downloading the models you need comes first. It is easiest to remember that installing the app and downloading models are two separate preparation steps.
With Ollama, start by confirming whether the model you selected actually runs on your computer, since cloud models are available too. Having installed the program is a different matter from having the AI run inside your machine.
Four things to check before you start
- Confirm that your OS and chip are supported by the program.
- Set aside space for model files separately from the program installation space.
- Look at the recommended memory and your chosen model’s memory requirements as two separate figures.
- If you plan to use it offline, download models in advance and confirm how they run.
Getting started with local AI means preparing a model to run on your own computer. Rather than reaching for a large model right away, it is better to do the work you need with a model that fits your machine. Once you are comfortable, you can weigh the memory requirements and runtime settings of the model you actually want to use.