How to Use a Local AI Model Offline With No Internet

By Austin Little

Most "AI" you've used lives in somebody else's building and stops the moment your Wi-Fi does. A local model is different: once the files are on your machine, the thinking happens on your machine. This guide is about the part nobody explains well, which is getting ready before you lose the connection so the tool still works after you do.

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

Who this is for

This is for people who want a writing and thinking helper that keeps working when the internet doesn't. Maybe you fly a lot. Maybe you spend weekends somewhere with one bar of signal. Maybe your neighborhood loses power and internet every winter storm. Maybe you simply don't want your drafts leaving the room.

Other Apiary guides cover setting up a local model on specific hardware: an old laptop, a gaming PC, a Windows machine without the command line, a Mac mini. This one is narrower and more practical. It assumes you're willing to install one free app, and it focuses on the question those guides skip: what do you need to have done before the connection goes away, and what changes once it has?

I'll keep it to free tools and free models. You won't need a subscription, an account, or a card for anything in here.

The one idea that makes offline AI make sense

A cloud chatbot is a window into someone else's computer. You type, your words travel across the internet, a big machine in a data center writes the reply, and the reply travels back. No internet, no reply.

A local model is a file. A big file, often several gigabytes, but a file. A local AI app loads that file into your computer's memory and does the math right there. Your words don't travel anywhere.

So the whole game of offline AI comes down to this: the internet is needed to get the file, not to use it. Download the app, download one or two models, test them while you still have a connection, and you've done the only part that needs Wi-Fi.

That's not my opinion; the app makers say it plainly. LM Studio's own documentation says it "can operate entirely offline, just make sure to get some model files first," and that chatting with models, chatting with documents, and running a local server don't require the internet. Ollama's FAQ says "Ollama runs locally" and that they don't see your prompts when you run locally.

What still works offline, and what doesn't

Being honest about the line saves frustration later.

Works without a connection

Chatting with a model you already downloaded. Ask questions, draft emails, rewrite paragraphs, brainstorm, make checklists.
Pasting in text you already have. Notes, a draft, an exported document, a long email you saved before you left.
Chatting with documents that live on your machine, in apps that support it. LM Studio's docs say documents you drag in for this stay on your machine and are processed locally.
Running a local server for other apps on the same computer, if you're the tinkering type. LM Studio's docs list this as an offline operation too.

Needs a connection

Finding and downloading new models. LM Studio's docs say model search and downloads need the internet. Ollama pulls models from the internet as well.
App updates. LM Studio's docs note that its in-app updater needs a network connection to check for new versions. Ollama on Mac and Windows downloads updates automatically when it can.
Anything "live." Today's weather, today's news, a store's opening hours, a current price, a recent law change. A local model has no window to the outside. It only knows what was baked into it during training, and that knowledge has a cutoff date.
Cloud features some apps bolt on. Ollama now offers optional cloud-hosted models and web search. Those need a connection, and they're a different privacy situation from local use.

The tricky middle

Some offline sessions fail not because of the model but because something else in the app wants the internet: a sign-in prompt, a model catalog that tries to refresh, a "check for updates" pop-up. These usually aren't fatal; they just look scary when you're on a plane. Practice once with Wi-Fi off at home so you know which pop-ups you can dismiss.

Pick your app: two free, beginner-friendly options

You don't need both. Pick one.

LM Studio: a desktop app with a chat window

LM Studio looks like a normal program. You open it, browse for a model, download it, and chat in a window. Its offline documentation is unusually clear about what does and doesn't need the internet, which is exactly what you want for this job. Its docs say it supports Mac, Windows, and Linux.

Good fit if: you want a chat window and buttons, and you'd rather not open a terminal.

Ollama: a small engine with an app and a command line

Ollama runs in the background and can be used through its own app or by typing commands. Its docs cover macOS, Windows, and Linux. It's popular with people who want other programs to talk to their local model.

There are other free local apps, and Apiary has separate comparisons. For offline use, the deciding factor is boring: pick the one you'll actually test before you leave.

Picking a model you can carry

This is where most offline plans go wrong. People download the most impressive-sounding model, then discover on the plane that it crawls or won't load at all.

Size matters more than the name

Local models come in sizes, usually described by "parameters" (a rough measure of how big the model's brain is) and by "quantization" (how much the file has been compressed). Apiary has a separate plain-English explainer on quantization. The short version for offline use:

Smaller models load faster, run faster, use less battery, and need less memory. They're less clever on hard questions.
Bigger models are smarter on hard questions but need much more memory and are slower, especially without a strong graphics chip.

I'm deliberately not giving you RAM numbers here. Whether a given model fits depends on your exact machine, the compression level, how long a conversation you keep going, and what else is open.

A practical rule

Download two models before you go offline:

A small, quick one for everyday tasks: rewording, short emails, summaries of text you paste in. This is your workhorse, and it's kind to your battery.
A medium one your machine can still handle for harder thinking: outlining something long, working through a tricky explanation. Use it when you're plugged in.

If your laptop struggles with the medium one at home, it will struggle more on battery on a tray table. Drop down a size.

Disk space

Model files are large. Before you download, check your free disk space. If you're tight, delete models you don't use. Ollama's FAQ lists where it stores model files on each operating system, and you can point it somewhere else, like an external drive, with a setting.

The pre-flight checklist (do this with the internet on)

Here is the routine I'd follow before any trip, outage season, or "I want this machine air-gapped" project.

1. Install and update the app

Install LM Studio or Ollama from its official site, not from a search ad or a look-alike download page. Then update it. You won't be able to update offline, so start from the current version.

2. Download your two models

Let them finish completely. A half-downloaded model is useless offline. Large downloads on hotel or airport Wi-Fi fail often, so do this at home.

3. Do a real test with Wi-Fi turned off

This is the step people skip. Turn off Wi-Fi (or unplug the network cable). Then:

Open the app from scratch.
Load each model.
Ask each one a real question.
Paste in a page of text and ask for a summary.
Close the app, reopen it, and do it again.

If anything fails, you've learned it at home instead of at 35,000 feet.

4. Gather your materials as files

A local model can only work with what you give it. Before you go, save what you'll need as plain files on your machine:

The draft you're working on.
Notes, outlines, reference material.
Important emails copied into a text document.
Any PDFs exported to text, if your app doesn't read PDFs directly.

Think of it like packing a bag. The model is the pen; you still need the paper.

5. Write a few saved prompts

When you're tired and offline, a ready-made instruction helps. Keep a small text file of prompts you like, such as:

"Rewrite this to be shorter and plainer. Keep my meaning. Don't add facts."
"Turn these messy notes into a numbered checklist."
"List questions I should ask before I make this decision. Don't decide for me."
"Point out sentences in this draft that are unclear."

6. Charge and plug in

Running a model works the processor hard, and that drains a battery faster than browsing does. Charge fully, bring the charger, and use the smaller model when you're on battery.

7. Turn off auto-sync surprises

If your computer syncs files to a cloud drive, decide what you want to happen when you reconnect. Offline drafts will upload when the connection returns. That's usually fine, but if you went offline specifically for privacy, keep those drafts in a folder that doesn't sync.

Using it once you're offline

Treat it like a sharp assistant with no phone

The model is good at language: wording, structure, summarizing, explaining general concepts, spotting gaps in an argument. It is bad at facts that change and facts it never saw. It can't check anything. It will sometimes state something wrong with complete confidence.

So offline, lean into the language jobs:

Tightening a draft.
Turning a brain-dump into an outline.
Writing three versions of a tricky email so you can pick one.
Explaining a concept you already roughly understand, so you can check it against your own knowledge.
Making a packing list, a to-do list, or a meeting agenda.

And be careful with the fact jobs:

Dates, prices, phone numbers, addresses, laws, medication doses, and anything recent. Mark these in your draft as "check later" and verify them when you're back online.

Keep conversations short

Local apps have a limit on how much text the model can "see" at once, often called the context window. Apiary has a separate explainer. Ollama's FAQ says its default context window is 4,096 tokens unless you change it. Very long conversations can make the model forget the start, slow down, or use more memory. Start a fresh chat for each new task, and paste in only what's relevant.

Save your work outside the chat

Copy anything good into your own document. Chat histories in local apps are stored on your machine, which is good for privacy but means a crash or a reinstall can lose them. Your document is the source of truth.

Offline for privacy, not just for travel

Some people go offline on purpose. The appeal is simple: if the machine isn't connected, your words can't leave it.

That's a real benefit, with limits worth knowing:

Offline protects what you type into the model. It doesn't encrypt your laptop, lock your screen, or protect files from someone who picks up your computer. Use your operating system's normal disk encryption and a password.
Reconnecting brings the world back. Sync tools, browser extensions, and update checks resume when you reconnect. If you're handling something sensitive, keep it in a non-synced folder.
The app itself still matters. Download from the official site. Local isn't automatically safe if the installer came from a sketchy mirror.
Cloud options inside local apps aren't local. If an app offers a "cloud model" or "web search" toggle, using it sends your prompt out. Decide on purpose.

For drafts that are genuinely sensitive (a medical situation, a legal dispute, a family matter), offline local AI is a reasonable place to organize your thoughts. It is not a place to get a decision.

AI explains, it doesn't decide

This matters even more offline, because you can't double-check anything in the moment.

If you're using a local model to think through something medical, legal, financial, electrical, or safety-related, use it to understand terms, organize notes, and write down questions. Then take those questions to a qualified professional: a doctor or pharmacist, a licensed attorney, a licensed electrician, a certified mechanic. A local model's training data is frozen, it can be wrong, and it has no idea about your specific situation.

A good offline habit: end those sessions with "List the questions I should ask a professional about this." That turns the model into a preparation tool, which is what it's good at.

When something goes wrong offline

The model won't load

Most often this is a memory problem. Close other programs, especially browsers with many tabs, and try again. If it still fails, switch to your smaller model. This is exactly why you brought two.

It's painfully slow

Same cause, usually. Smaller model, fewer open programs, and plug in if you can. Some laptops slow the processor on battery to save power.

The app keeps trying to connect

Dismiss update or sign-in prompts. If a catalog page won't load, that's expected offline; go to your already-downloaded models instead. LM Studio's docs specifically note that the model catalog and search need a connection, so a blank catalog page offline is normal.

The answers are nonsense

Small models sometimes ramble or repeat. Try a fresh chat, a shorter and clearer instruction, or the bigger model when you're plugged in. Asking for a specific format ("give me five bullet points") often helps.

You forgot to download something

There's no fix offline. Write a note for next time, and work with what you have.

A sample offline day

Here's how this might look on a long flight or a weekend at a cabin with no signal.

Morning, plugged in: Open your medium model. Paste in your messy notes for a project. Ask for an outline. Edit the outline yourself until it's right.

Midday, on battery: Switch to the small model. Draft one section at a time in your own words, then ask the model to point out unclear sentences. Accept some suggestions, reject others.

Afternoon: Paste a long saved email thread. Ask for a summary and a list of open questions. Draft a reply. Mark any dates or numbers as "verify."

Evening, back online: Check every "verify" item against a real source. Send nothing you haven't read twice.

That's a productive day with no subscription, no account, and nothing leaving the laptop.

Preparing a household "outage kit" machine

If your area loses internet in storms, you might set up one family computer with a local model ready to go. A few suggestions:

Keep one small model downloaded and tested.
Put a shortcut on the desktop with a plain name like "Offline Helper."
Write a one-page instruction sheet: how to open it, how to start a new chat, and the warning that it can't know current news or emergency information.
Next to it, keep printed emergency contacts and official sources.

A local model can help a bored teenager with homework during an outage or help you draft an insurance note afterward. It cannot tell you whether a road is open or a downed line is live. Stay away from downed power lines and call the utility or emergency services.

Common questions

Is offline AI as good as the big cloud chatbots?

Usually not on the hardest questions, and it has no live information. For everyday writing and organizing, a well-chosen local model is often good enough, and some people prefer it because nothing leaves the machine.

Do I need a powerful computer?

You need enough memory for the model you choose. Many people start on the computer they already have with a small model. Apiary's other guides cover specific machines. Test before you rely on it.

Can I use offline AI on my phone?

Some phones now run small AI features on the device, and there are apps that run small models locally. Capabilities vary a lot by phone and app, and app stores have plenty of look-alikes.

Does offline mean nobody can see my prompts?

If the model runs locally and you aren't using a cloud feature, your prompts stay on your machine, according to both LM Studio's and Ollama's documentation. Anyone with access to your computer can still see your files and chat history, so lock your computer.

Can I move a model to another computer without the internet?

Model files can be copied, and LM Studio's docs mention "sideloading" models obtained outside the app. The process differs by app, so look up your app's instructions while you're still online and test it first.

Will the model know what happened last week?

No. A local model only knows what was in its training data up to its cutoff. Ask it about recent events and it may guess. Treat anything time-sensitive as "verify later."

The short version

Install one free local app from its official site and update it.
Download a small model and a medium model while you have good internet.
Turn off Wi-Fi at home and test both. Really test.
Save the files you'll want to work on.
Offline, use it for language jobs: drafting, rewording, outlining, summarizing.
Mark facts, numbers, and anything time-sensitive to verify later.
For medical, legal, electrical, or safety questions, use it to prepare questions, then talk to a qualified professional.

The internet gets you the file. After that, the helper is yours, and it works wherever your laptop does.

Sources

LM Studio Docs, "Offline Operation": https://lmstudio.ai/docs/app/offline (fetched 2026-10-04)
Ollama FAQ: https://docs.ollama.com/faq (fetched 2026-10-04)