claudevm.bundle: the hidden 10.8 GB Ubuntu VM in Claude Desktop

The Surprise of 10 GB

You install Claude Desktop on your Mac. Everything seems fine; the app doesn’t take up much space. But one day, you check your disk and find this:

~/Library/Application Support/Claude/vm_bundles/claudevm.bundle

10.8 GB.

Yes: one single file called claudevm.bundle, tucked away in Claude’s vm_bundles folder, taking up almost eleven gigabytes of your disk space. Ten gigabytes for a chatbot? What does it contain— the extended edition of The Lord of the Rings trilogy?

Not quite. claudevm.bundle contains Ubuntu.

The Holy Trinity of Claude Products

Before getting into the what, let’s cover the why. Anthropic offers three ways to use Claude:

The website is the straightforward option: everything happens on Anthropic’s cloud, and nothing touches your machine. But if you want Claude to perform tasks locally on your Mac — creating documents, running code, managing files — you’ll need something more.

That’s where Cowork and Claude Code come in, offering two completely different philosophies for local execution.

Cowork: Claude Code for the Rest of Us

Cowork is the name of the agent Anthropic released on January 12, 2026. Its official tagline:

“Claude Code for Everything Else You Do”

While Claude Code is designed for developers working in the terminal, Cowork is aimed at users who aren’t coding. Think: the lawyer organizing contracts, the analyst extracting data from 50 PDFs, the teacher turning class notes into a presentation.

What It Can Do

• You grant it access to a folder on your Mac.
• Claude can read, edit, and create files within that folder.
• It can generate Office documents (Word, Excel, PowerPoint) without requiring additional software.
• It can run Python or JavaScript code if necessary.
• It can launch multiple “sub-agents” simultaneously for heavy tasks.

Real-World Example

“Hey Claude, I’ve got 30 receipt screenshots in this folder. Make me a spreadsheet with the date, description, and amount from each one.”

And Claude delivers. It reads the images, extracts the data using OCR, generates an Excel file, and saves it right there in the folder.

Fun Fact

Cowork was almost entirely built by Claude Code in just 1.5 weeks. Claude developing the next generation of Claude products—think Inception, without Leonardo DiCaprio but with a lot more tokens.

The VM: A Security Trick

Here’s the clever bit. When Cowork runs code or works with files, it doesn’t interact with your Mac directly. Instead, it operates inside a completely isolated Ubuntu virtual machine.

Why? Because letting an AI execute arbitrary code on your system is like handing the keys to your house to a brilliant but unpredictable stranger. They might be trustworthy, but why risk giving them access to the kitchen knives?

What’s Inside Those 10 GB

In simpler terms: a complete office environment, neatly sandboxed.

The Security Model

┌─────────────────────────────────────────┐
│            Your Mac (host)              │
│  ┌───────────────────────────────────┐  │
│  │      Ubuntu VM (sandbox)          │  │
│  │  ┌─────────────────────────────┐  │  │
│  │  │   Claude runs here          │  │  │
│  │  │   - No access to your disk  │  │  │
│  │  │   - Filtered network        │  │  │
│  │  │   - Resets every session    │  │  │
│  │  └─────────────────────────────┘  │  │
│  │                                   │  │
│  │  /mnt/user-data/ ← Your folder   │  │
│  └───────────────────────────────────┘  │
│                                         │
│  The rest of your Mac: INACCESSIBLE     │
└─────────────────────────────────────────┘

The VM can only access the folder you’ve given it. It cannot touch your photos, documents, or .ssh keys. Even its network access is restricted — it can only connect to specific services like PyPI, npm, or GitHub.

If Claude makes a mistake inside the VM, the damage is contained there.

Claude Code: No Safety Net

Then there’s Claude Code, which is… a different beast altogether.

brew install claude-code
claude

Claude Code skips the VM. No sandbox, no isolation, no safety net. When you ask it to run a command, it executes directly in your terminal — with your permissions, access to your files, and your SSH keys.

It’s the difference between showing someone a photo of your house and giving them your house keys.

Why Would You Want This?

Because as a developer, you often need Claude to have real access:

• Your Git repository with all its files.
• Your build tools (npm, cargo, uv, etc.).
• Your tests to execute them.
• Your credentials to handle deployments.

Developing software inside an isolated VM that knows nothing about your project? Not practical.

Permission Model

Claude Code isn’t a free-for-all. It uses a permission system:

• Always Allowed: Reading files, searching through code.
• Requires Your Approval: Running commands, writing files, making network requests.

Each time it wants to execute a command, it asks for your confirmation:

Claude wants to execute: npm test
[Approve] [Deny] [Always Approve]

The model is simple: you trust Claude to read and understand your code, but you control what it does.

Cowork vs. Claude Code: The Comparison

When to Use Each

Use Cowork when:

• You’re creating or editing Office documents.
• You need to run someone else’s code (and don’t fully trust it).
• You’re organizing personal files.
• You’re not a developer.

Use Claude Code when:

• You’re developing software.
• You need access to your repo, tests, and tools.
• You’re handling deployments.
• You’re debugging something requiring full system access.

And What About Copilot?

Many people ask how this compares to GitHub Copilot. Briefly: they serve different purposes.

Copilot is AI-assisted autocomplete. It lives in your IDE and suggests code as you type, line by line. It’s fast, intuitive, and great for boilerplate.

Claude (Cowork/Code) is an AI agent. It plans, executes complete tasks, and works across multiple files. It’s slower but much more powerful.

The consensus among users:

“Copilot makes you faster. Claude makes you better. In 2026, you need both.”

Can You Delete the VM?

Yes. If you don’t use code execution features:

1. Open Claude Desktop → Settings.
2. Disable “Code Execution and File Creation.”
3. Optionally: rm -rf ~/Library/Application\ Support/Claude/vm_bundles/

This will free up 10 GB, and you’ll still have regular chat functionality with Claude.

Final Thoughts

The next time someone asks why Claude Desktop takes up so much space, you’ll have the answer: it runs a full Ubuntu environment with LibreOffice, Chromium, Python, Node.js, and everything needed to handle real file tasks securely.

It’s the price of safety. Letting Claude run code directly on your Mac would mean risking real damage every time the AI makes a mistake (and yes, large language models make mistakes).

I’d rather give up 10 GB to a VM than have an accidental rm -rf / wipe out my home directory.

Want to learn more about Claude Code? Check out my git optimization tutorial that I co-wrote with it.

This article was originally published in Spanish and translated with the help of AI.