Mint’s Developer on Safe Mac Cleanup — and Why Trust Beats “Junk Found”

Most Mac cleaners lead with how much “junk” they found. Mint focuses on something else: clear categories, safer actions, and ways to reclaim space without making deletion the only option.

That approach comes from Yukun Gao of DZG Studio, a former data scientist who built Mint after getting tired of sorting thousands of files with no practical way to decide what still mattered. After reviewing the app, we spoke with Yukun about Mint’s design choices, AI agent clutter, and why trust matters more than free space.

TheSweetBits: What frustration with existing Mac cleaners made you decide to build Mint instead of just using—or complaining about—the alternatives?

Yukun: “I’ve always appreciated things that are simple and end in an action. Before Mint I was a data scientist, and to me data and models matter most when they turn into something you can act on.

Looking at my own disk, there was simply too much to take in: thousands of files, and no easy way to know what each one was or whether it still mattered. Working that out myself took a lot of time.

So I wanted Mint to be clear and safe, the two things you picked up on in your review: clearly actionable, without too many technical details. It’s the balance I’ve found so far. There may be better ones later, but it’s the product I want to build today.

Talking with users, I kept hearing one thing: “Knowing I won’t delete the wrong thing matters more to me than how much space I get back.” That made me step back. Deleting is only one lever. What other levers can give a Mac its space back? That question became a series of experiments, and the ones that worked became treatments: first Optimize, space back without deleting anything; then what’s safe to clean; then what’s yours to decide.

Some complexity is unavoidable. When it is, I’d rather solve it systematically, so it doesn’t land on you. That’s what all those experiments are for.

Organizing came from the same place. Opening your Mac shouldn’t start with half an hour of cleanup. Mint takes care of that, so you go straight to the work.”

How do you decide the boundary between “Safe to clean” and “Yours”? What’s the hardest type of file or folder to classify correctly?

“Nothing is Safe to clean because a rule says so. We study each kind of file on real Macs, including what happens after it’s removed. A label has to be proven, not assumed. If Mint can’t prove it, it doesn’t guess: the item goes to Yours, where you decide.

The hardest cases are files that are technically safe and still not worth deleting. On my own Mac, Mint deleted the same node_modules folder 11 times in three days, and it came back every time. Nothing was lost, and nothing was gained. We caught it in Mint’s own records, fixed it, and wrote it up publicly.

Many rules take dozens of hours of research before they ship, some well over a hundred.”

Mint lets AI agents interact with storage through its CLI and MCP server. How do you make agent-driven file operations safe, and what should users still be responsible for?

“Mint is the brain; the agent is the hand.

We’ve all seen the posts: an agent wiped someone’s files, or a worktree with unsaved work. The agent wasn’t careless. It lacked context, and context is what Mint has.

So an agent can ask Mint anything, but it acts only through Mint’s checks. It works from a plan Mint made, never from a raw path, and the plan expires in 15 minutes. Mint checks each item again before it moves. Agents can only move items to the Trash, only what Mint rated safe, and never delete permanently. Anything that needs a judgment call comes back to a person. We even tell the agent: a plan is not user consent.

What stays with users: Mint only guards what goes through Mint, so your agent’s own permissions still matter. And what’s worth keeping is always your call.”

Also, AI coding agents are creating a new kind of storage problem. What have you learned from observing how tools like Claude Code, Cursor, and Codex use disk space?

“Two things: the pictures, and the leftovers.

Agents check their work by looking at it, and every screenshot is saved in the conversation. On my Mac, one Codex store was 22.83 GB, and 71% of it was images, many of them the same screenshot again and again. 92% of those repeats were in conversations used in the past six days, so deleting old chats barely helped.

The leftovers are harder. The agent can’t decide for you, and you often don’t know what it left behind. It’s like a middle schooler asking a high schooler to cook dinner: dinner’s great, everyone’s happy, and nobody mentions the kitchen.

AI produces output faster than anyone can review it, so “delete this, delete that” won’t keep up. In the long run we need a systematic way to evaluate and solve for what agents produce and leave behind.”

Who is the ideal Mint user right now — power users with messy Downloads folders, people running lots of AI tools, or someone else?

“Two kinds of people. Mac power users who push their Mac every day, especially developers and anyone running AI tools. And anyone whose Mac is a year or two in, technical or not: take a look, there’s usually something to gain. With Optimize alone, many of our friends got a decent amount of space back without deleting anything.”

Is there a lesser-known or “hidden” part of Mint that you wish more people discovered?

“Auto Care, and the idea behind it.

People think of it as “do the chore on a schedule.” I think of it as renovating a home. What matters isn’t the area on paper; it’s how much of it you can actually live in. A Mac is the same. Optimize raises that usable share without taking anything away, and Auto Care keeps it there every day while the Mac is idle.

The other one is Memory. Mint shows at a glance which apps sit idle holding memory. Friends open it and say, “I had no idea that app used so much.” That’s often why their Mac stalled.

Next, I’d like to build a community around Mint, where people share what they find.”

Outside of Mint, what’s something you’re building or exploring that people might not expect?

“Mint keeps me more than busy for now. I’ve explored voice, though there are good open-source options there already. What I keep coming back to is workflow: tools that take friction out of the day. When something new is ready, it’ll be at DZG Studio first.”

Closing thoughts

Yukun’s answers reinforce what stood out in our own testing: Mint is less interested in dramatic cleanup totals than in decisions users can trust. The four-category model, the cautious approach to “Safe to clean,” and the agent safety rules all point in the same direction — reclaim space without turning cleanup into a gamble.

The AI angle is especially timely. As tools like Claude Code, Cursor, and Codex become part of daily Mac work, they also create a new class of storage problems: repeated screenshots, abandoned worktrees, and leftovers users may not even know exist. Mint’s CLI and MCP support are an early attempt to give agents better context while keeping irreversible actions out of their hands.

If you want a Mac cleaner that prioritizes clarity over aggressive deletion, Mint is worth a closer look. You can learn more at mintstorage.app, and follow other work from the studio at dzgapp.com.

Also read: Our full Mint review

Loved the article, share!