How to Run Local AI on Windows Without the Command Line

By Austin Little

"Local AI" sounds like something for people with three monitors and a hoodie. It doesn't have to be. On a reasonably capable Windows PC, you can install an app, click a model, and chat with it, with your words staying on your own machine and no monthly bill. Here's how, using only what each tool's official documentation says.

AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events.

What "local AI" means, in one paragraph

A cloud chatbot runs on a company's servers. Everything you type travels over the internet to them. A local AI runs a language model on your own computer. You download the model file once, and after that the conversation can happen entirely on your machine, even with the Wi-Fi off. The trade-off is that your PC does the work, so what you can run, and how fast, depends on your hardware. In exchange you get privacy, no subscription, and no usage caps set by someone else.

Why bother?

A few honest reasons people choose local:

• Privacy. LM Studio's offline documentation says that once a model is on your machine, nothing you enter when chatting leaves your device. Jan's Windows install page says it stores your data locally and that nothing is sent to the cloud.
• No subscription.
• Works offline. Great on a plane, in a cabin, or anywhere with patchy internet.
• Sensitive drafts. Medical questions you're preparing for a doctor, a family budget, a personal letter, a home inventory: things you'd rather not paste into a website.

And the honest reasons not to bother: if your PC is older or low on memory, local models may be slow or limited, and the biggest, smartest models usually still live in the cloud. Local AI is a great tool, not a magic one.

Will my PC handle it?

I'm only going to give you numbers the app makers publish themselves. They're recommendations and minimums, not guarantees of a good experience, and actual performance depends heavily on which model you pick.

What the official docs say

LM Studio (system requirements), for Windows:

• Supported on both x64 and ARM (Snapdragon X Elite) systems.
• CPU: AVX2 instruction set support is required (for x64).
• RAM: "LLMs can consume a lot of RAM. At least 16GB of RAM is recommended."
• GPU: "at least 4GB of dedicated VRAM is recommended."

Jan (Windows installation):

• Windows 10 or higher.
• CPU: AVX2 support required. Jan's page names Intel Haswell (2013+) and AMD Excavator (2015+) as examples.
• Memory: 8GB minimum (16GB recommended).
• GPU: 6GB VRAM minimum for NVIDIA, AMD, or Intel Arc GPUs, for GPU acceleration.
• Storage: 10GB free space minimum.

Ollama (Windows docs):

• Windows 10 22H2 or newer, Home or Pro.
• Native Windows app with NVIDIA and AMD Radeon GPU support. The docs list minimum driver requirements for each, so check them if you have a graphics card.
• At least 4GB of disk space for the app itself, plus more for models, which the docs say "can be tens to hundreds of GB in size."
• The installer doesn't require Administrator rights and installs in your home folder by default.

How to check your own PC (no commands)

• RAM and Windows version: Open Settings → System → About.
• Graphics card and its memory: Open Task Manager (right-click the Start button), go to the Performance tab, and click GPU.
• Free disk space: Open File Explorer → This PC.
• AVX2 support: Nearly all modern desktop and laptop CPUs from the last decade have it, but "nearly" isn't "all." Search your exact CPU model on the manufacturer's site. Jan's page gives Intel Haswell (2013+) and AMD Excavator (2015+) as reference points.

Don't have a dedicated graphics card? The apps can still run models on your CPU. It's just usually slower. Start with the smallest models and see how it feels. I won't promise a speed, because it depends on too many things.

The three apps at a glance

You don't need all three. Pick one. If you want my nudge:

• Want it to "just work" with the fewest decisions? Try Jan, since it downloads a default model on first launch.
• Want to browse lots of models and see details? Try LM Studio.
• Heard of Ollama, or want something that also works with other tools later? Try the Ollama app.

Option 1: LM Studio

LM Studio is a desktop app for downloading and running open models. According to its July 2025 blog post, it's free to use at home and at work, with no form or separate commercial license needed. Its download page says the app is available under its terms of use.

Install

1. Go to the official site: lmstudio.ai/download.
2. Download LM Studio for Windows. The page also offers other products, including one called "LM Studio Bionic" and a command-line server tool called llmster.
3. Run the installer like any other Windows program.

Only download from the official website. Lookalike sites and "free AI" bundles are a common way to get malware.

Download your first model

LM Studio's getting started guide describes the flow:

1. Open the Discover tab.
2. Pick one of the curated options, or search for a model by name.
3. Download it.

Its docs mention model families like Qwen, Mistral, Gemma, and gpt-oss as examples. If you're unsure, start small. Smaller models download faster, use less memory, and are much more forgiving on modest hardware. LM Studio shows details for each model.

Load it and chat

From the same guide:

1. Go to the Chat tab.
2. Open the model loader.
3. Select the model you downloaded.
4. Optionally choose load settings. You can leave the defaults.

LM Studio explains that "loading a model" means allocating memory to hold the model in your computer's RAM. That's why RAM matters, and why closing other big programs can help.

Then type. That's it.

Chat with documents

LM Studio's offline page says you can drag and drop a document into the app to chat with it, and that the document stays on your machine and all processing is done locally. That makes it a good fit for "summarize this long PDF" jobs you don't want to upload anywhere.

What needs internet and what doesn't

Per LM Studio's offline documentation:

• Doesn't need internet: chatting with downloaded models, chatting with documents, running a local server.
• Needs internet: searching for models, downloading models, the Discover tab's model stats, downloading runtimes, and checking for app updates.

So: download while online, then chat as offline as you like.

Option 2: Jan

Jan describes itself on its homepage as an open-source ChatGPT replacement. Its quickstart is short because the app tries to do most of the setup for you.

Install

From Jan's Windows installation page:

1. Download Jan from the official site.
2. Run the downloaded .exe file.
3. Wait for installation to finish.
4. Launch Jan.

First launch

Jan's quickstart says Jan automatically downloads its default model on first launch, and once the download completes you're ready to chat with no extra setup. Then just type in the message box.

More models: the Hub

Per the quickstart:

1. Open the Hub from the left sidebar.
2. Browse or search for a model, then click Download.
3. Once it's downloaded, pick it from the model selector in any chat.

Jan says each model in the Hub shows whether it fits your hardware, which is especially helpful if you're not sure what your PC can handle.

Keep it local: watch these settings

Jan can do both local and cloud. Its quickstart says local models are the default, running entirely on your machine, privately, offline, and with no API key. But it also mentions:

• Cloud model providers (like OpenAI, Anthropic, and Google) that you can connect with an API key under Settings → Model Providers.
• Built-in web search and fetch tools for current events.

If privacy is your reason for going local, don't add cloud API keys, and be aware that web search sends your search queries out to the internet.

Where your data lives

Jan's Windows page says it stores everything locally, including downloaded models, conversation threads, settings, and logs, in a data folder under your user's AppData\Roaming\Jan\data. If you want to back up your chats or free up disk space, that's where to look.

GPU acceleration (optional)

If you have a supported graphics card, Jan's Windows page says to go to Settings → Hardware → GPUs and turn on the switch if it isn't already on. The page also lists driver requirements for NVIDIA cards. Updating graphics drivers through the manufacturer's normal installer doesn't need the command line.

Option 3: The Ollama Windows app

Many people know Ollama as a command-line tool, and its docs do describe the ollama command. But Ollama announced in July 2025 that its macOS and Windows apps now include a way to download and chat with models in a regular window, plus file drag and drop for text and PDFs and image input for models that support it.

Install

Ollama's Windows docs say the easiest way to install on Windows is the OllamaSetup.exe installer, which installs into your account without needing Administrator rights. After installing, Ollama runs in the background.

One heads-up: when I checked Ollama's Windows download page today, the most prominent install option was a one-line PowerShell command, which is exactly the command line we're avoiding. The docs still describe the .exe installer.

Chat

Once it's installed, open Ollama from the Start menu or the system tray. In the app, choose a model, let it download, and start typing. Per Ollama's announcement, you can drag a text file or PDF into the window to ask about it.

Local vs. Ollama's cloud

Ollama offers both. Its download page describes two ways to run models: locally on your computer, where "speed depends on the hardware," and in Ollama's cloud, where models run on Ollama's servers. If privacy and no-cost are your goals, stick with local models and don't sign in for cloud models.

Change where models are stored, still without commands

Models can be large. Ollama's docs say they "can be tens to hundreds of GB in size." If your C: drive is tight, the Windows docs describe a point-and-click way to move model storage:

1. Open Settings (Windows 11) or Control Panel (Windows 10) and search for "environment variables."
2. Click Edit environment variables for your account.
3. Create a new variable named OLLAMA_MODELS with the folder where you want models stored, for example a folder on a bigger drive.
4. Click OK or Apply.
5. Quit Ollama from the system tray and relaunch it from the Start menu.

It's a little technical, but it's all clicking and typing in boxes. No terminal.

Uninstall

Per the docs, Ollama registers an uninstaller under Add or remove programs in Windows Settings. If you moved the models folder, the uninstaller won't delete those models, so remove that folder yourself if you want the space back.

Your first ten minutes, whichever app you pick

1. Install from the official site.
2. Download one small model. Resist the urge to grab the biggest one.
3. Ask something easy. "Explain what a 401(k) is in plain English, in five sentences."
4. Ask something useful. "Rewrite this email to be shorter and friendlier: [paste]."
5. Turn off Wi-Fi and ask again. This is the fun part. It still works. That's local.
6. Notice the speed. If it's painfully slow, try a smaller model before deciding local AI isn't for you.

Picking a model without a computer science degree

Model names look like license plates. Here's how to read them without needing benchmarks:

• Size is usually in the name as a number followed by "B" (billions of parameters). Smaller numbers mean smaller, lighter models. Larger numbers generally mean more capable but much more demanding.
• Start small, move up. If a small model runs comfortably and you want better answers, try the next size up. If your PC starts struggling, step back down.
• Use the app's guidance. Jan says its Hub shows whether a model fits your hardware. LM Studio shows model details before download.
• Instruction or "instruct" or "chat" variants are tuned for conversation, which is what most people want.

I'm deliberately not saying "model X needs Y gigabytes," because it depends on the model, its file format, and settings.

Privacy: how private is "local," really?

Local is very private for the chat itself, with a few honest caveats:

• Downloads go over the internet. Searching for and downloading models contacts the app's servers or model hosts. LM Studio's docs, for example, mention requests to huggingface.co when searching.
• Update checks go online. LM Studio's offline page says the app checks for updates when you open it.
• Optional cloud features aren't local. Jan's cloud providers and web search, and Ollama's cloud models, send data off your machine. Leave them off for maximum privacy.
• Your PC's own security still matters. If someone else uses your Windows account, they can open your chat history. Use a password on your Windows account.

If you want the strictest setup: download your model, then use the app with Wi-Fi off.

Things local AI is great for

• Drafting and rewriting emails, letters, and cover letters.
• Summarizing long documents you'd rather not upload.
• Brainstorming gift ideas, meal plans, and trip outlines.
• Explaining jargon from a bill, a medical portal, or a contract so you can ask a professional better questions.
• Organizing messy notes into lists or tables, like a home inventory or a recipe collection.

Things to keep in human hands

• Medical, legal, and financial decisions. Use AI to prepare questions, not to decide.
• Facts you'll repeat to others. Local models can be wrong, and smaller ones may be wrong more often. Check anything important against a reliable source.
• Current events. A local model only knows what it was trained on, unless you've connected a web tool, which reduces privacy.

Troubleshooting without the command line

"It's really slow." Try a smaller model. Close other heavy programs, like games, video editors, and dozens of browser tabs. If you have a graphics card, check the app's GPU setting (Jan: Settings → Hardware → GPUs).

"The model won't load" or "out of memory." The model is probably too big for your RAM or VRAM. Pick a smaller one.

"It installed, but I can't find it." Check the Start menu. For Ollama, look for its icon in the system tray near the clock. It runs in the background.

"Ollama shows strange squares." Ollama's docs say progress characters may show as squares in some older Windows 10 terminal fonts. That only affects the command-line view, which you're not using.

"I'm running out of disk space." Models are big. Delete ones you don't use from within the app, or move Ollama's model storage as described above. Jan keeps its data under AppData\Roaming\Jan\data.

"My antivirus complained." Make sure you downloaded from the official site. If you did and you're still unsure, don't override your security software. Ask someone you trust to take a look.

Uninstalling cleanly

• Ollama: Settings → Add or remove programs. Delete a custom models folder separately if you made one.
• Jan: Its Windows page describes uninstalling from Control Panel's Programs section, then deleting the Jan folder in C:\Users\[username]\AppData\Roaming to remove leftover data. Its docs warn that deleted data can't be recovered, so back up anything you want first.

Common myths

"You need to be a programmer." Not for these three apps. Install, click, chat.

"You need an expensive gaming PC." A better PC helps, but the apps can run smaller models on ordinary hardware. Check the official requirements above and start small.

"Local AI is as good as the best cloud AI." Usually not for the largest tasks. But for drafting, summarizing, and explaining, small local models can be genuinely useful, and they're private.

"Free means there's a catch." These apps publish their terms. LM Studio publicly announced it's free at home and work, and Jan describes itself as open source. Ollama sells cloud plans, and its local app is free to download. Read the terms if you're curious. You don't have to pay to run local models.

Quick answers

"How do I run AI locally on Windows without coding?" Install LM Studio, Jan, or the Ollama Windows app from their official sites, download a model inside the app, and chat.

"Which is easiest?" Jan downloads a default model on first launch, so it has the fewest steps. LM Studio is great for browsing models. Ollama's app is simple and lives in your system tray.

"How much RAM do I need for local AI?" LM Studio recommends at least 16GB on Windows. Jan lists 8GB minimum and 16GB recommended. Smaller models are more forgiving.

"Does local AI work offline?" Yes, once the model is downloaded. LM Studio's docs say chatting with downloaded models doesn't need the internet.

"Is local AI private?" The chat stays on your machine. Model downloads, update checks, and any optional cloud or web features do use the internet.

Bottom line

You don't need a terminal, a subscription, or a computer science degree to run AI on your own Windows PC. Pick one of LM Studio, Jan, or the Ollama app, install it from the official site, download a small model, and start chatting. Keep cloud features off if privacy is the point, start small if your PC is modest, and keep important decisions in human hands.

References

• LM Studio — System Requirements: https://lmstudio.ai/docs/app/system-requirements (fetched 2026-10-04)
• LM Studio — Get started: https://lmstudio.ai/docs/app/basics (fetched 2026-10-04)
• LM Studio — Offline Operation: https://lmstudio.ai/docs/app/offline (fetched 2026-10-04)
• LM Studio — Free to use at work: https://lmstudio.ai/blog/free-for-work (fetched 2026-10-04)
• LM Studio — Download: https://lmstudio.ai/download (fetched 2026-10-04)
• Jan — Homepage: https://www.jan.ai/ (fetched 2026-10-04)
• Jan — Windows Installation: https://www.jan.ai/docs/desktop/install/windows (fetched 2026-10-04)
• Jan — Quickstart: https://www.jan.ai/docs/desktop/quickstart (fetched 2026-10-04)
• Ollama — Windows: https://docs.ollama.com/windows (fetched 2026-10-04)
• Ollama — New app announcement: https://ollama.com/blog/new-app (fetched 2026-10-04)
• Ollama — Download for Windows: https://ollama.com/download/windows (fetched 2026-10-04)