How to use Claude to free up space on your Windows PC


AI assistants are quietly becoming the Swiss Army knife no one expected. I’ve already turned ChatGPT into a project management system and a Grammarly Pro replacement. But Claude is different—it has access to my file system. So when I noticed my storage was nearing 80%, I decided to use Claude to free up some disk space.

How does Claude know what files you have on your PC?

Two words: Claude Cowork

Claude Cowork home screen with Chat Cowork and Code tabs

If you’ve only used Claude through the web or mobile app, it might be surprising to learn that it can read and interact with your system files. However, this is a standard feature in the Claude desktop app when you switch to Cowork mode. It lets you mount any folder or directory on your computer, allowing Claude to find files, read them, create new ones, move them, and yes—even delete them.

For security, it’s generally recommended to sandbox Claude to a specific folder so it can only access that directory. That said, I’ve been using Claude Cowork since its release and haven’t experienced any issues—even with broader access. So when my storage usage crept close to 80%, I pointed it at my C drive to identify large, unnecessary files and help remove them.


A Pixel 10 with Claude open and an iPad Air with  an Obsidian second brain open.


How Claude fixed my messy Obsidian vault in 5 minutes (prompts included)

Your second brain has become a second junk drawer. Claude can fix that.

The simplest safeguard is to explicitly tell Claude not to delete anything without confirmation. It will also prompt you for permission before making significant changes. Some users click “Allow” without reading, but you should always review what it’s asking for. As long as you do that, there’s little risk of Claude deleting anything unintentionally.

How I use Claude to find and visualize my files

It’s all about the right prompts

If you simply ask Claude what’s taking up space on your PC, it will run a basic scan and return a generic summary. That’s not useless—but it’s not especially helpful either. The better approach is to give it a structured prompt that clearly defines what to look for and how to present the results.

This is the prompt I used:

Scan my PC for files that are wasting space. I want a thorough, multi-pass scan — not just a surface-level check.

**IMPORTANT: DO NOT DELETE ANYTHING. ONLY SCAN AND SHOW ME**

Specifically, look for:
- Temporary and cache files (Windows Temp, AppData caches, browser caches)
- Large files over 500MB (any location)
- Old installer files (.exe, .msi, .iso in Downloads or elsewhere)
- Duplicate files (same name + size in multiple locations)
- Leftover folders from uninstalled software
- Log files that have grown out of control
- Thumbnails and icon caches

Once the scan is complete, present the findings as an interactive visual dashboard — use a treemap along with a pie chart to show which categories and folders are consuming the most space. Break it down by: total space used, recoverable space (safe to delete), and space that needs my review before deleting.

For each item or category flagged, give me a brief reason why it's junk — or flag it if it might actually be useful, so I can make the call.

Again, **DO NOT DELETE ANYTHING**. Show me everything first.

The most important part is the instruction: “do not delete anything.” In fact, it’s so important I’ve mentioned it twice. You want to review what Claude found before you let it touch anything. After that, you can either ask Claude to remove specific files or handle the cleanup yourself if you prefer more control.

The request for a visual dashboard—specifically a treemap and pie chart—also makes a significant difference. It helps you quickly understand where your storage is going, rather than sorting through a long list of files.

You can customize this prompt further by adding new scan targets or changing how the data is presented.

You can make the process more hands-on if you prefer

Not everyone is comfortable giving Claude direct access to their file system. The main concern is accidental deletion. There’s also the fact that Claude can technically read your files, although in practice, if the task is focused on file sizes, it isn’t going to inspect your documents. Still, if you’d rather keep it at arm’s length, there’s an alternative that avoids giving Claude any file access at all.

The idea is simple: instead of letting Claude scan your PC directly, you ask it to generate PowerShell commands that you run yourself. You then paste the output back into Claude, which analyzes the results and guides you from there. It’s a bit more hands-on, but significantly safer.

Side-by-side view of Claude Cowork analyzing C drive and PowerShell showing disk usage at 343 GB used.-1

Here’s the prompt for that approach:

I want to find what's wasting space on my PC, but I'd rather not give you direct access to my file system. Instead, give me a series of PowerShell commands to run one at a time. I'll paste the output back to you after each one.

Start with a broad scan, then we'll drill down based on what we find.

Cover these areas:
- Top 20 largest files on C:\\
- Windows Temp and AppData\\Local\\Temp folder sizes
- Downloads folder — sorted by size
- Any .log files over 50MB
- Installer files (.exe, .msi, .iso) outside of Program Files
- Browser cache folders (Chrome, Edge, Firefox — whichever exist)

After I share each output, analyze it and tell me: what's safe to delete, what needs my review, and how much space I'd recover. Once we've gone through all the scans, give me a summary of total recoverable space broken down by category.

Don't give me all the commands at once — go one step at a time so we can work through this together.


A hand holding some screenshots of Claude 'Projects' and the Claude logo in the center.


Claude isn’t just for programmers: Here’s 7 ways I use it in my everyday life

Vibe coding isn’t all you can do.

Is Claude better than a dedicated disk cleaner?

In some ways, yes—in most ways, no

Claude Cowork all findings table listing AI ML models, Claude VM files, Windows Installer Cache, and package data flagged for review with reasons.-1

If you’re on Windows, you already have Storage Sense built in. On the third-party side, tools like BleachBit are genuinely powerful and cover a wide range of cleanup tasks. Both options are free and purpose-built for cleaning junk from your PC.

The Claude workflow, by comparison, consumes tokens. On a free plan, that’s a clear limitation. Even with Claude Pro, it’s still not “free” in the same way traditional disk cleaners are. But is that extra cost worth it?

The main advantage Claude has over standard disk cleaners is context awareness. When BleachBit encounters a file, it evaluates whether it’s junk based on predefined rules. Claude, on the other hand, evaluates whether it’s junk for you. That’s a fundamentally different question.

For example, during my scan, Claude flagged 40 GB of LLM models stored in an LM Studio folder. BleachBit would likely ignore it because it doesn’t fall into a typical junk category. But Claude recognized that I exclusively run local models through Ollama. I hadn’t used LM Studio in months, so it correctly identified the entire folder as a candidate for deletion. That’s the kind of judgment call a generic tool isn’t designed to make.


Computer running Windows 11.


Stop installing these 4 apps—Windows does it all now

It’s all included in the box now.


Claude vs. BleachBit or Storage Sense

Claude isn’t a replacement for BleachBit or Storage Sense—it’s a complement. If you want automated cleanup based on well-defined rules, traditional tools still do that best. But if you want something that understands your setup and can make informed decisions about what’s actually worth keeping, Claude has a clear edge.



Source link

Leave a Reply

Subscribe to Our Newsletter

Get our latest articles delivered straight to your inbox. No spam, we promise.

Recent Reviews


Vibe coding has taken the development world by storm—and it truly is a modern marvel to behold. The problem is, the vibe coding rush is going to leave a lot of apps broken in its wake once people move on to the next craze. At the end of the day, many of us are going to be left with apps that are broken with no fixes in sight.

A lot of vibe “coders” are really just prompt typers

And they’ve never touched a line of code

An AI robot using a computer with a prompt field on the screen. Credit: Lucas Gouveia / How-To Geek

Vibe coding made development available to the masses like never before. You can simply take an AI tool, type a prompt into a text box, and out pops an app. It probably needs some refinement, but, typically, version one is still functional whenever you’re vibe coding.

The problem comes from “developers” who have never written a line of code. They’re just using vibe coding because it’s cool or they think they can make a quick buck, but they really have no knowledge of development—or any desire to learn proper development.

Think of those types of vibe coders as people who realize they can use a calculator and online tools to solve math problems for them, so they try to build a rocket. They might be able to make something work in some way, but they’ll never reach the moon, even though they think they can.

Anyone can vibe code a prototype

But you really need to know what you’re doing to build for the long haul

For those who don’t know what they’re doing, vibe coding is a fantastic way to build a prototype. I’ve vibe coded several projects so far, and out of everything I’ve done, I’ve realized one thing—vibe coding is only as good as the person behind the keyboard. I have spent more time debugging the fruits of my vibe coding than I have actually vibe coding.

Each project that I’ve built with vibe coding could have easily been “viable” within an hour or two, sometimes even less time than that. But, to make something of actual quality, it has always taken many, many hours.

Vibe coding is definitely faster than traditional coding if you’re a one-man team, but it’s not something that is fast by any means if you’re after a quality product. The same goes for continued updates.

I’ve spent the better part of three months building a weather app for iPhone. It’s a simple app, but it also has quite a lot of complex things going on in the background.

It recently got released in the App Store—no small feat at all. But, I still get a few crash reports a week, and I’m constantly squashing bugs and working on new features for the app. This is because I’m planning on supporting the app for a long time, not just the weekend I released it, and that takes a lot more work.

Vibe coders often jump from app to app without thinking of longevity

The app was a weekend project, after all

A relaxed man lounging on an orange beanbag watches as a friendly yellow robot works on a laptop for him, while multiple red exclamation-mark warning icons float around them. Credit: Lucas Gouveia/How-To Geek | ViDI Studio/Shutterstock

I’ve seen it far too often, a vibe coder touting that they built this “complex app” in 48 hours, as if that is something to be celebrated. Sure, it’s cool that a working version of an app was up and running in two days, but how well does it work? How many bugs are still in it? Are there race conditions that cause a random crash?

My weather app has a weird race condition right now I’m tracking down. It crashes, on occasion, when opened from Spotlight on an iPhone. Not every time does that cause a crash, just sometimes.

If a vibe coder’s only goal is to build apps in short amounts of time so they can brag about how fast they built the app, they likely aren’t going to take the time to fix little things like that.

I don’t vibe code my apps that way, and I know many other vibe coders that aren’t that way—but we all started with actual coding, not typing a prompt.


Anyone can be a vibe coder, but not all vibe coders are developers

“And when everyone’s super… no one will be.” – Syndrome, The Incredibles. It might be from a kids’ movie, but it rings true in the era of vibe coding. When everyone thinks they can build an app in a weekend, everyone thinks they’re a developer.

By contrast, not every vibe coder is actually a developer, and that’s the problem. It’s hard to know if the app you’re using was built by someone who has plans to support the app long-term or not—and that’s why there’s going to be a lot of broken apps in the future.

I can see it now, the apps that people built in a weekend as a challenge will simply go without updates. While the app might work for the first few weeks or months just fine, an API update comes along and breaks the app’s compatibility. It’s at that point we’ll see who was vibe coding to build an app versus who was vibe coding just for online clout—and the sad part is, consumers will lose out more often than not with broken apps.



Source link