How to Back Up Your Local AI Models and Chats Before a Laptop Dies

By Austin Little

Running AI on your own laptop means your models, your chats, and your settings live on one drive you own. That's the whole point, and it's also the risk. Here's where the files actually sit, what's worth saving, and a free backup routine you can set up in an afternoon.

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

Short answer (read this first)

If your laptop is making a bad noise right now, do this before anything else:

1. Copy your chats first. Conversations are small and impossible to re-download. Models are big and you can usually pull them again.
2. For LM Studio, copy the conversations folder inside .lmstudio in your home folder. LM Studio's docs give ~/.lmstudio/conversations/ on Mac and Linux and %USERPROFILE%\.lmstudio\conversations on Windows.
3. For Open WebUI in Docker, your chats, users, and settings live in the Docker volume (named open-webui in the official quick start). Open WebUI's docs give a one-line tar command to save it to a file. It's further down this page.
4. For Ollama, the models live in ~/.ollama/models on macOS, /usr/share/ollama/.ollama/models on Linux with the standard installer, and C:\Users\%username%\.ollama\models on Windows, per Ollama's FAQ. Copy that folder if you have the room and the time. If you don't, write down the model names (ollama list) so you can pull them again later.
5. Copy everything to a second physical place, like an external drive, another computer, or a USB stick for the small stuff. A second folder on the same dying drive isn't a backup.

That's the emergency version. The rest of this guide is the calm version, where you set it up once and stop worrying.

Why local AI needs its own backup plan

When you use a cloud chatbot, the company holds your history. That has real privacy costs, but it does mean a coffee spill doesn't erase your conversations. When you run AI locally with Ollama, LM Studio, or Open WebUI, you've taken that job back. Nobody else has a copy unless you made one.

A lot of people find this out too late, and the way they find out is boring. A battery swells. A drive stops mounting. An operating system update goes sideways and the repair shop wipes the disk "to be safe." The laptop gets left on a train. None of these are rare, and none of them care how careful you were with your prompts.

There's a second, quieter reason. Local AI folders tend to get big and messy. You pull a model to try it, forget it, pull three more, and soon the models folder is the largest thing on your computer. When disk space runs low, people delete things in a hurry, and sometimes the wrong folder goes. A backup routine forces you to look at what's there and decide what matters.

The three kinds of files, and which ones matter most

Before you copy anything, it helps to sort your local AI stuff into three piles. They have very different value.

1. Your conversations and notes (irreplaceable, small)

This is your chat history: the drafts you worked through, the explanations you asked for, the system prompts you tuned. Nobody can give these back to you. They're also usually small text or database files. This pile always gets backed up.

2. Your settings and custom work (hard to recreate, small)

This covers things like:

• Custom Modelfiles you wrote for Ollama (your own system prompt, temperature, and so on).
• Presets and per-model settings in LM Studio.
• In Open WebUI, your users, admin settings, connections, uploaded documents, and knowledge bases, which all live in the same data volume as the chats.
• Any notes you keep about which model you use for what.

You could rebuild these from memory. You won't enjoy it. Back these up too.

3. The model files themselves (replaceable in most cases, huge)

Model weights are the big files, often several gigabytes each and sometimes far more. Ollama's own Windows docs warn that models "can be tens to hundreds of GB in size." In most cases you downloaded them from a public library (Ollama's library or Hugging Face) and can download them again.

So model files are the lowest priority, with two big exceptions:

• You made or changed the model yourself, for example by importing a GGUF file you converted or quantized, or by building a custom model with ollama create. Back up the source files and the Modelfile.
• The download might not be there later. Publishers sometimes take down or rename model repos.

Where Ollama keeps your stuff

Ollama is the tool most Apiary readers start with, so we'll begin there. Everything in this section comes from Ollama's official FAQ and Windows docs, checked October 1, 2026.

Ollama model folder

Ollama's FAQ lists the default model locations as:

• macOS: ~/.ollama/models
• Linux: /usr/share/ollama/.ollama/models
• Windows: C:\Users\%username%\.ollama\models

Ollama's Windows docs also say %HOMEPATH%\.ollama "contains models and configuration," and that %LOCALAPPDATA%\Ollama holds logs and downloaded updates. You don't need to back up the logs or updates.

If you moved your models, Ollama uses the OLLAMA_MODELS environment variable to point at a different folder. If you (or someone helping you) set that, the models are wherever it points, not in the default spot. On Linux with the standard installer, Ollama's docs note that the ollama user needs read and write access to a custom folder.

Your Ollama key

Ollama's FAQ also documents a public key file, id_ed25519.pub, in the same .ollama folder (~/.ollama/ on macOS, /usr/share/ollama/.ollama/ on Linux, C:\Users\<you>\.ollama\ on Windows). It's used to push models and to talk to ollama.com for cloud features. Next to it sits the private half of the key pair. If you only ever run local models, you can let a new machine make a new key. If you've pushed models under your ollama.com account, treat the private key like a password.

Ollama chat history

This is where it gets murky. Ollama started as a command-line tool, and the command line doesn't keep a full chat archive the way a chat app does.

Until that's confirmed, the safe move is simple. If a chat in the Ollama app matters, copy or export it out of the app into a plain text or Markdown file in a folder you already back up. If you used Ollama mostly through another front end (Open WebUI, an editor plugin, or something else), your chats live in that tool, not in Ollama.

Saving the list of what you have

Even if you skip copying model files, keep a list. In a terminal:

ollama list > my-ollama-models.txt

That writes the names of your installed models to a small text file. On a new machine you can pull each one again by name. If you built custom models, also save the Modelfile you used.

Where LM Studio keeps your stuff

LM Studio's docs are clear on both models and chats. Checked October 1, 2026.

LM Studio models

LM Studio's "Import Models" page shows the expected models directory as ~/.lmstudio/models/, organized like this:

~/.lmstudio/models/
└── publisher/
    └── model/
        └── model-file.gguf

You can change that folder in the app. LM Studio's docs say to head to My Models. Its command-line tool, lms, follows whatever directory you set there.

LM Studio chats

LM Studio's "Manage chats" page gives the conversations folder as:

• Mac / Linux: ~/.lmstudio/conversations/
• Windows: %USERPROFILE%\.lmstudio\conversations

The docs also say you can right-click a chat and choose "Reveal in Finder" or "Show in File Explorer" to jump straight to it. Conversations are stored as JSON files. LM Studio warns that it's "NOT recommended to edit them manually, nor to rely on their structure." That matters for backups. Copy the files as they are; don't try to tidy or merge them. If you want a human-readable copy of an important chat, export or paste it into a text file as well.

Folders you made in the chat sidebar are part of how LM Studio organizes that conversations folder. Copy the whole folder, not just the chats you remember.

Where Open WebUI keeps your stuff

Open WebUI is a browser-based chat front end that many people run on top of Ollama. It's also the easiest of the three to back up, because its docs spell the process out. Checked October 1, 2026.

If you run Open WebUI in Docker (the common setup)

Open WebUI's quick start uses -v open-webui:/app/backend/data, and describes that volume this way: "Your chats, users and settings live in this volume. It survives updates; never run without it."

Its updating guide adds that the volume holds the database (chats, users, settings, admin configuration), uploaded files (documents, images, knowledge base content), and generated content.

The official backup command, from the same guide, run from the folder where you want the backup file to land:

docker run --rm -v open-webui:/data -v $(pwd):/backup \
  alpine tar czf /backup/openwebui-$(date +%Y%m%d).tar.gz /data

In plain English: start a tiny throwaway container, attach your Open WebUI data, and pack it into a dated .tar.gz file in your current folder. Open WebUI recommends backing up before every update and on a regular schedule.

The restore command in the same guide deletes everything in the volume before it extracts the backup. Their docs say so in a caution box, and we'll repeat it: double-check you're restoring the right file to the right machine.

If you used Docker Compose, your volume name may have a project prefix. Open WebUI's uninstall notes refer to it as <your_project_name>_open-webui. Run docker volume ls to see the real name, and use that in the command.

If you run Open WebUI with Python (pip, uv, conda)

Open WebUI's docs say to set DATA_DIR to choose where your data lives, for example DATA_DIR=~/.open-webui open-webui serve. If you set it, back up that folder.

If you didn't set it, the data sits inside the Python environment, at site-packages/open_webui/data. That's fragile. Removing or rebuilding the environment can take your chats with it, and the docs warn that uvx "might store it in a temporary folder that gets deleted when the process ends." If you're on a Python install with no DATA_DIR, fixing that is the first thing to do, before any backup routine.

Don't forget the secret key

Open WebUI uses WEBUI_SECRET_KEY to sign logins. Its docs say a new key signs everyone out, and that Python installs write the key to a .webui_secret_key file in the folder you start from. Losing it isn't a disaster (you just log in again), but if you restore onto a new machine, keeping the same key avoids confusion. Store it the way you'd store a password: in a password manager, not in a text file next to the backup.

A free backup routine that actually works

You don't need a paid backup service for this. You need two things: a second place to put files, and a habit.

Pick your second place

Free or already-owned options, roughly from simplest up:

• An external USB drive. The best home for model files, because they're big. Keep it unplugged when you're not backing up, so the same accident or malware can't hit both copies.
• Another computer in the house. A desktop, an old laptop, or a family member's machine with space to spare. Copy over the home network or with a drive.
• A USB stick. Fine for chats and settings, which are small. Usually too small for models.
• A free cloud storage tier you already have, for chats and settings only.

On privacy: the reason many people run AI locally is to keep their chats off other people's servers. Uploading those chats to a cloud drive undoes part of that. If you do use cloud storage, think about encrypting the backup file first (more on that below), or keep chats on physical drives only.

The order of operations

Here's a routine you can do on the first of every month, or before any big update:

1. Close the apps. Quit LM Studio and the Ollama app, and stop Open WebUI (docker stop open-webui, or stop the Python server). Copying a database while it's being written can give you a broken copy.
2. Back up chats and settings first. LM Studio's conversations folder, Open WebUI's volume or DATA_DIR, your Modelfiles, your notes.
3. Save your model list. ollama list > my-ollama-models.txt, plus a screenshot or note of LM Studio's My Models page.
4. Copy model folders if they changed and you have space. A big copy only needs to happen when you've added something you care about.
5. Name things with dates. For example, ai-backup-2026-10-01. Future you will thank you.
6. Check the copy (see the next section).

Copy commands you can use

On Mac or Linux, a simple, resumable copy to an external drive:

rsync -a --progress ~/.lmstudio/conversations/ /Volumes/BackupDrive/ai-backup/lmstudio-conversations/

Change /Volumes/BackupDrive to wherever your drive mounts. On Linux that's often under /media/ or /mnt/. rsync only copies what changed, so the second run is fast. That makes it a good fit for a big models folder.

For Ollama's Linux models folder, which is owned by the ollama system user, you may need sudo to read it:

sudo rsync -a --progress /usr/share/ollama/.ollama/models/ /media/you/BackupDrive/ai-backup/ollama-models/

On Windows, File Explorer drag-and-drop works fine for small folders. For big model folders, the built-in robocopy command handles interruptions better:

robocopy "%USERPROFILE%\.ollama\models" "E:\ai-backup\ollama-models" /E

Check that your backup is real

A backup you've never opened is a hope, not a backup. Checking takes five minutes:

• Open the backup folder and make sure the files are there, with sizes that look right. An empty folder or a 0-byte file means something went wrong.
• For Open WebUI, list what's inside the archive without restoring it: tar tzf openwebui-20261001.tar.gz | head. You should see paths under data/.
• For LM Studio chats, open one JSON file in a text editor and read a line or two. Don't save changes. Just confirm it's your conversation.
• For model files, compare the folder size on the laptop to the copy. If you want to be thorough, tools like shasum (Mac/Linux) or certutil -hashfile (Windows) can confirm a big file copied exactly. That's optional.

Once a year, or whenever you get a new machine, do a restore drill. Restore onto a spare computer or a fresh user account and confirm your chats show up. It's the only way to know the whole chain works.

Encrypting the backup (optional, recommended for chats)

Your chats may hold things you'd never post publicly: health questions, money worries, work drafts, family business. If the backup drive gets lost, whoever finds it can read them.

Free ways to protect them:

• Use your operating system's built-in disk encryption on the external drive, if your system offers it.
• Put the chats in an encrypted archive. Free tools like 7-Zip (Windows) can create password-protected archives with encryption.
• Use a password manager to store whatever password you choose.

When the laptop is already failing

If you're reading this with a laptop that's freezing, clicking, or barely booting, change your priorities:

1. Stop installing or downloading anything. Every write to a failing drive is a risk.
2. Grab chats and settings first. They're small and fast to copy, and they can't be replaced.
3. Skip the models unless they're custom. Write down the model names instead. You can pull them again on a healthy machine.
4. If it won't boot, don't keep power-cycling it. A repair shop or a knowledgeable friend can often pull the drive and read it from another computer. Tell them which folders you need, using the paths in this guide, so they don't have to guess.
5. If a shop is going to wipe the drive, ask them to copy your home folder first, including hidden folders that start with a dot, like .lmstudio and .ollama. Those are hidden by default on Mac and Linux, and people miss them.

Hidden folders are the number one way local AI backups fail. On a Mac, you can show hidden files in Finder with Command-Shift-Period.

Restoring on a new machine

When you get a new laptop, put things back in this order:

1. Install the apps fresh from their official sites (ollama.com, lmstudio.ai, Open WebUI's docs). Don't copy program files from the old machine.
2. Restore chats and settings. Put LM Studio's conversations folder back in .lmstudio. For Open WebUI, create the volume and use the restore steps in their updating guide, or point DATA_DIR at your restored folder.
3. Restore or re-pull models. Either copy your saved models folder back to the default location (or point OLLAMA_MODELS or LM Studio's My Models setting at it), or pull the models again using your saved list.
4. Open a few chats to confirm everything came across.

A note on versions: Open WebUI's docs warn that database migrations are one-way. If you restore an old backup into a newer Open WebUI, it will usually upgrade the data. Going the other way, a newer backup into an older version, may not work. When in doubt, install the same or a newer version than the one you backed up from.

Moving models to an external drive on purpose

Some people back up model files simply by keeping them on an external drive all the time. Both main tools support this:

• Ollama: set OLLAMA_MODELS to a folder on the drive. Ollama's FAQ covers setting environment variables on each OS: launchctl setenv on Mac, systemctl edit ollama.service on Linux, and the Environment Variables settings on Windows. Restart Ollama afterward.
• LM Studio: change the models directory in My Models.

The trade-off is that the drive has to be plugged in whenever you want to use those models, and a slow drive can make models load slowly. It also means you now have one copy on the external drive instead of one on the laptop. That's still one copy.

A one-page checklist

Print this or paste it into your notes.

Every month (or before updates):

• [ ] Quit Ollama app, LM Studio, and stop Open WebUI
• [ ] Copy ~/.lmstudio/conversations/ (or %USERPROFILE%\.lmstudio\conversations)
• [ ] Run the Open WebUI volume backup command, or copy your DATA_DIR
• [ ] Export any Ollama-app chats you care about to text
• [ ] Save ollama list output and any custom Modelfiles
• [ ] Copy changed model folders to the external drive (optional)
• [ ] Open the backup and spot-check it
• [ ] Unplug the backup drive and put it somewhere else

Once a year:

• [ ] Restore drill on a spare machine or a fresh user account
• [ ] Delete models you no longer use, from the laptop and the backup
• [ ] Check that your backup passwords still work

What this guide won't tell you

We won't tell you that you need a paid backup service. You don't. We won't tell you how many gigabytes your models take up, because it depends entirely on which models and versions you downloaded. And we won't publish a file path we couldn't confirm in the tool's own docs today. Where we couldn't, we've said so.

Local AI is a good deal: your prompts stay with you, nothing is metered, and nobody can switch it off on you. The cost is that you're the IT department. With a monthly habit and one external drive, it's a small job.

FAQ

Do I need to back up the model files? Usually not, as long as you saved the model names and the models are still available to download. Back them up if you made or changed them yourself, if you rely on one exact version, or if your internet is slow or capped.

Where does Ollama keep its models? Per Ollama's FAQ: ~/.ollama/models on macOS, /usr/share/ollama/.ollama/models on Linux (standard installer), and C:\Users\%username%\.ollama\models on Windows, unless OLLAMA_MODELS points somewhere else.

Where does LM Studio keep my chats? Per LM Studio's docs: ~/.lmstudio/conversations/ on Mac and Linux, %USERPROFILE%\.lmstudio\conversations on Windows. You can also right-click a chat and use "Reveal in Finder" or "Show in File Explorer."

How do I back up Open WebUI? For Docker, back up the open-webui volume with the tar command from Open WebUI's updating guide. For Python installs, back up the folder set in DATA_DIR. If you never set one, set one first.

Can I just copy my whole home folder? Yes, and that's a fine catch-all, as long as your copy tool includes hidden folders (the ones starting with a dot). On Linux, remember Ollama's default model folder is under /usr/share/ollama, not your home folder.

Is it safe to put my AI chats in cloud storage? It's a trade-off. It protects against a dead laptop but hands a copy to the storage company. If you do it, encrypt the backup first.

Sources

Checked October 1, 2026:

• Ollama FAQ: https://docs.ollama.com/faq
• Ollama on Windows: https://docs.ollama.com/windows
• Ollama, Importing a Model: https://docs.ollama.com/import
• LM Studio, Manage chats: https://lmstudio.ai/docs/app/basics/chat
• LM Studio, Import Models: https://lmstudio.ai/docs/app/advanced/import-model
• LM Studio, Download an LLM: https://lmstudio.ai/docs/app/basics/download-model
• Open WebUI Quick Start: https://docs.openwebui.com/getting-started/quick-start/
• Open WebUI, Updating (Backup & Restore): https://docs.openwebui.com/getting-started/updating/