C盘管家 (disk-butler): a 2.4 MB Tauri disk and memory helper for Windows users who do not want a scare screen

The problem C盘管家 targets: cleanup tools that frighten instead of inform

The README opens with a scene rather than a feature list. A C: drive fills up, a cleaning utility announces thousands of junk items and an urgent threat, and the user, now alarmed, accepts a bundle of extra software. The alternative the project names is the expert route: hand someone WizTree and a directory tree full of English paths, and they still will not know whether WinSxS is safe to touch.

C盘管家 positions itself between those two. The stated intent is to treat the user as an adult: no scare copy, no bundled installers, and for every operation an answer to three questions, what is this, why does it take space, and what happens if I delete it. The audience is explicitly non-expert Windows users, and the interface language is Chinese. The design principles section in the README lists five commitments, including a ban on fear-based wording and a rule that every number on screen carries its source, so a label reading scanned 5 minutes ago means five minutes ago rather than being silently refreshed.

That framing is the product. A cleaner that deletes the wrong thing is worse than one that deletes nothing, and the README's own comparison table makes the deletion boundary the headline difference rather than scan speed or feature count.

How the scan pipeline works: MFT direct read, a read-only service, and a jwalk fallback

The architecture is a Tauri 2 shell: a Rust backend plus the system WebView2 for the interface, with React 18, TypeScript, Vite and Tailwind CSS v4 on the front end. The installer is 2.4 MB, which the README attributes to using the platform webview instead of bundling a browser runtime.

The scanning engine is the part worth understanding. According to the README, it reads the NTFS master file table directly through the ntfs-reader crate, the same general approach the README attributes to WizTree, so a whole-disk tally does not require walking every directory. When the MFT path is unavailable, the project falls back to parallel traversal with jwalk, which is slower.

What makes the fast path practical without administrator rights is a separate read-only scanning service. The README describes it as modeled on Everything: the service does the privileged work, the main program talks to it over named-pipe IPC, and the UI gets sub-second scans without elevation. The service does not write to disk and does not use the network. If it is disabled or removed by security software, the application detects the absence and reverts to the slow scan, and the README states there is a one-click repair to restore the fast path.

Visualization uses d3-hierarchy with a squarify layout, drawn by React, with framer-motion for transitions. Colors carry category meaning (system, software, cache, personal files), and a knowledge base of more than 60 Chinese software ecosystem entries maps known directories to plain-language descriptions. Scan results are stored locally and reloaded on the next launch.

Installing C盘管家 and running a first scan

The README points at the Releases page and names the artifact DiskButler-Setup.exe. There is no package manager path documented, no winget or scoop command, so a download from GitHub is the intended route. The README also lists a mirror link for users in China, described as an accelerated download that opens a page with a start button.

System requirements are narrow and stated plainly: Windows 10 or 11, 64-bit. WebView2 is expected to be present already, so no extra runtime is installed. Windows 7 and 8 cannot run it, and the README says the installer reports this rather than installing something that will not start.

Double-click the installer and it completes quickly at 2.4 MB. On first launch, SmartScreen may warn about an unknown publisher, because the README states the project has no code signing certificate. The documented workaround is the standard More info, then Run anyway. That is a real friction point for a tool aimed at non-experts, and the README answers it by pointing at the source.

If you prefer to build it yourself, the README gives this sequence:

git clone https://github.com/YGTQ3/disk-butler.git
cd disk-butler
npm install
npm run tauri dev

The prerequisites are Node.js 18 or newer, a Rust MSVC toolchain, and VS Build Tools 2022 with the C++ workload. Development mode works without the scan service, but then you are on the slow path. To try the fast path while developing, the README documents building the service and starting it from an administrator terminal:

npm run build:svc
src-tauri\target\release\disk-butler-svc.exe run-console

For a release build, the README recommends capping parallelism on a 16 GB machine:

set CARGO_BUILD_JOBS=2 && npm run tauri build

Once it is running, the first useful action is a disk scan from the disk insight view. You should see a colored treemap where large blocks are immediately visible and clicking a block descends into it. The README says the previous result reappears on the next launch instead of rescanning.

The three-tier cleanup model and why the middle tier is the interesting one

Cleanup items are split into three groups with different defaults. The first is junk residue such as temporary files and leftover update packages, checked by default. The third is anything containing personal data, such as the recycle bin and partial downloads, and the README states these are never checked by default. The middle tier is performance caches: npm, browser caches, IDE indexes. These are also not checked by default, and the README's justification is unusually candid for this category of software: caches speed up the applications that own them, so when disk space is not tight the tool advises leaving them alone.

A cleaner that argues against cleaning is a deliberate design position, and it only works if the user can check the claim. Each item expands to a detail view that lists the concrete paths to be removed, and the README asserts that what the interface shows is what gets executed, enforced by a whitelist in the backend rather than by front-end filtering. That is the strongest technical claim in the document, because it means a bug in the UI cannot widen the deletion scope.

The deep cleanup view handles Windows component store backups, the WinSxS directory that the README notes can run to tens of gigabytes. This does not use a custom deletion routine. It shells out to Microsoft's DISM in two stages: a read-only analysis that leaves the original report untouched and reports how much can be freed and whether Microsoft recommends it, then execution only after explicit confirmation. The README frames this as sharing lineage with the built-in Disk Cleanup tool.

Startup management and memory inspection round out the feature set. Startup entries are labeled with their real path, current memory use, and a grouped suggestion (disable, your call, keep). Toggles are reversible. Memory inspection ranks processes by usage and marks which ones can be exited.

Where C盘管家 is the wrong tool, and where the documentation goes quiet

The platform boundary is the first hard limit. Windows 10 and 11 x64 only. No macOS, no Linux, and no Windows 7 or 8. If your problem is disk space on a Mac or a Linux server, this project has nothing to offer, and the README does not pretend otherwise.

The second limit is structural. The fast scan depends on two things at once: an NTFS volume and a running privileged service. The README's own description of the fallback means a user whose security software removes the service without warning will experience a sudden, unexplained slowdown until they find the repair action. The MFT read also implies NTFS specifically, so a drive formatted with another filesystem will not get the fast path. The README does not document behavior for exFAT or ReFS volumes, and it does not document rollback for cleanup operations beyond the statement that personal-data items are unchecked by default.

There is also an asymmetry in the deep cleanup section. DISM is a Microsoft tool with its own constraints, and the README describes the two-step analysis and confirmation flow but says nothing about what happens if DISM fails midway, or how long component cleanup takes on a heavily used machine.

Finally, the code signing gap is a genuine adoption obstacle rather than a cosmetic one. A tool whose entire pitch is that you can trust what it deletes asks users to click through a SmartScreen warning on first run. The README's answer, that the source is available for audit or self-build, is honest but places the burden on the user. For a non-technical audience, that is the weakest link in the whole design.

How C盘管家 differs from WizTree and from bundled cleanup suites

The README names WizTree directly as the source of the MFT-reading idea, and the comparison is instructive because the two tools solve different halves of the problem. WizTree answers where the space went. It presents a directory tree and expects you to interpret it. C盘管家 answers the same question but adds the interpretation layer: the color categories, the knowledge base of known Chinese software directories, and per-directory plain-language notes about what the folder is and whether it can be touched. The scan mechanism is similar in spirit; the delivery is aimed at a different reader.

The contrast with bundled cleanup suites is sharper and the README makes it explicitly in a table. The dimensions chosen are installer size, network behavior, advertising, background services, deletion logic, and source availability. On network behavior the claim is absolute: zero requests, nothing uploaded, no telemetry, no update check. On background services the claim is one read-only scanner that can be disabled from the service manager, against what the README describes as multiple services that restart each other in the bundled-suite model.

A reader should weigh those claims as claims. The zero-network behavior is verifiable with a firewall rule or a packet capture, and the deletion whitelist is verifiable in the Rust source. The service count is verifiable in the service manager. These are all checkable by a moderately technical user, which is presumably the point of publishing them.

Licence, build cost, and what upgrading looks like

The project is MIT licensed, which permits use, modification and redistribution provided the copyright notice and permission notice are retained. There is no separate commercial tier, no dual licence, and no contributor licence agreement mentioned in the README. For a tool that deletes files, the practical implication is that anyone can fork it and ship a modified build under the same terms, including a build with different deletion rules. The licence does not guarantee that the binary you downloaded matches the published source, which is why the README's audit-or-self-build line matters more here than it would for a library.

Build cost is documented and non-trivial. You need Node.js 18 or newer, a Rust MSVC toolchain, and VS Build Tools 2022 with the C++ workload. The README flags a specific failure: LLVM ERROR: out of memory, which it attributes to the large windows binding crate. The stated mitigations are an opt-level = 1 override for that crate in Cargo.toml, limiting parallelism with CARGO_BUILD_JOBS=1 or 2, and ensuring a sufficient page file. On a 16 GB machine the README suggests CARGO_BUILD_JOBS=2. None of that is unusual for a Tauri project, but it is a real hour-or-two commitment rather than a two-minute install.

Upgrade cost for end users is low. Releases are frequent, with v0.7.7, v0.7.8 and v0.7.9 all published in August 2026, and the last push to the repository was on 2026-09-17. The README states there is no update check, so upgrading means returning to the releases page and downloading a new installer. The repository carries per-version release notes files from 0.4.0 through 0.7.8, so the change history is inspectable before you upgrade. The roadmap lists automated GitHub Actions builds as an unchecked item, which means release artifacts are currently produced by hand.

Editorial conclusion

Adopt C盘管家 if you are a Windows 10 or 11 x64 user who wants to see exactly which paths a cleaner will delete before it deletes them, and if you accept that WinSxS cleanup goes through Microsoft's DISM rather than a custom engine. Do not adopt it if you run Windows 7 or 8, if you need macOS or Linux, or if you want a tool that decides for you. Verify three things first: that the release page carries a DiskButler-Setup.exe for the version you want, that your machine has WebView2 present as the README assumes, and that the read-only scan service installs and stays enabled, because disabling it drops you to the slower jwalk fallback. If you build from source instead, confirm your page file is large enough before running npm run tauri build, since the README documents LLVM out-of-memory failures on 16 GB machines.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Releases
5. YGTQ3/disk-butler on GitHub