Clean Mind
v1.2.1
Abstract — A free, open-source disk analyzer that doesn't just chart your disk — it tells you what's safe to delete, why, and the exact command that rebuilds it. It goes deepest on a developer's disk (node_modules, build caches, .venvs).
Storage fills up quietly — and the agentic-coding era made it worse. A few weeks of leaning on coding agents, and the disk-full warning arrives. You didn't download anything big. You'd just been building. Every cleaner shows you where the space went, then leaves you to guess whether deleting a folder will wreck a project. Clean Mind actually understands your disk.
fig. 02a — SCAN
A parallel Rust scanner walks any folder — or drag one in. Sizes are true on-disk sizes: hardlink-aware, APFS-clone-aware. A full home directory — about 1.2 million files — scans in ~8 seconds.
fig. 02b — UNDERSTAND
The scan becomes a treemap you can read — every tile sized by the space it takes. Drill down with breadcrumbs, switch to a sortable list, or search the entire scan.
fig. 02c — CLASSIFY
A deterministic rules engine recognizes developer artifacts — node_modules beside a package.json, cargo target/, DerivedData — and records how each one regenerates. The rules engine, not the AI, is the source of truth.
fig. 02d — CLEAN
One click sends verified-safe items to the OS Trash — recoverable, never automatic. Permanent deletion sits behind a type-to-confirm gate.
+0 MB reclaimed → Trash
Everything the scan finds lands in one of three tiers. The AI is never trusted on its own.
Confirmed by the rules engine — like node_modules next to a package.json. Fully reproducible, rebuild command included.
AI suggestions that no rule verifies. Always shown with reasoning — never one-click, never deleted automatically.
A hard-coded denylist — documents, photos, .ssh, system paths — that neither the rules nor the AI can override.
Everything runs on your machine. The optional AI pass is the only thing that can leave it — and it's built so it can't leak your files. With pseudonymization on, it never even learns your folder names:
Structural names like node_modules stay readable. Mapping answers back to real paths happens locally.
No telemetry, no account, no background service. Scan results live only in memory — fresh every launch.
Strictly opt-in. Bring your own Anthropic or OpenAI-compatible key, or run a local model with Ollama.
The AI sees directory names, sizes and ages — never file contents.
API keys are stored in the OS keychain, not in config files.
The DaisyDisk idea — free, cross-platform, and it actually understands a developer's disk.
One native app on macOS, Linux and Windows — x64 and arm64, each with the right OS conventions (Trash vs Recycle Bin, per-platform volumes, long paths on Windows).
12+ · universal · Apple silicon & Intel
recommended — homebrew
brew install --cask MS-Teja/clean-mind/clean-mind
or the DMG: CleanMind-1.2.1-macos.dmg
Not notarized — first launch on macOS 15+: open once, then System Settings → Privacy & Security → Open Anyway (earlier versions: right-click → Open). Grant Full Disk Access for complete results.
10 & 11 · x64 & arm64
recommended — scoop
scoop bucket add clean-mind https://github.com/MS-Teja/scoop-clean-mind
scoop install clean-mind
or portable zips: x64 (Intel/AMD) · arm64 (Snapdragon X)
Extract and run clean_mind.exe. If SmartScreen warns: More info → Run anyway.
GTK 3 · amd64 & arm64
recommended — one-line installer
curl -fsSL https://ms-teja.github.io/clean-mind/install.sh | sh
Picks the right build for your CPU: the .deb on Debian/Ubuntu/Kali/Mint, a per-user install elsewhere (no root).
or the .deb:
amd64 ·
arm64
— then sudo apt install ./<file>.deb
or the tarball + install.sh (user-level, no root):
x64 ·
arm64
every build + checksums: github.com/MS-Teja/clean-mind/releases