Windows 10/11 x64 · v0.2.0-rc.2

The uninstaller that
shows its work.

Remove the app, find what it left behind, clear the caches your AI tools quietly filled, and see what your PC is actually doing — before anything changes. One calm utility instead of five noisy ones.

Review-first · recoverable cleanup · AI & IDE cache review · no telemetry · no RAM-booster claims

Cyron Performance page showing a 20-second spike capture with peak CPU, kernel and disk figures and the process using the most CPU.
Scroll to explore
Review before every change Quarantine before deletion Read-only diagnostics by default Zero telemetry by architecture

Review first

Nothing moves until you say so.

Cyron builds a server-held plan, shows the exact files and registry keys, and executes only after you confirm. Cleanup is recoverable by default.

Exact targets

Every command, path and registry key is listed before the plan becomes usable.

Recoverable

Files move to a journaled quarantine with no-overwrite restore, not silent deletion.

Single-use plans

Confirmations expire, cannot be replayed, and are revalidated against the live system.

Local only

No account, no analytics, no network client. Your inventory and traces stay on the PC.

01 Measure

The scan reads the filesystem and reports what exists.

02 Review

Exact paths, risk and expected effect appear in a plan.

03 Quarantine

An approved run journals each move before it happens.

04 Restore

Put a batch back at any time; existing files are never overwritten.

One utility

Five jobs, one clear point of view.

Uninstall, explore, clean, diagnose and recover without handing your PC to a black box. Scroll through the work.

01

Uninstall without leftovers

Win32, Store, Steam and portable apps in one inventory. Batch runs only what has a registered vendor uninstaller, and leftover files are reviewed separately.

  • Batch queue refuses Store, Steam and unregistered entries
  • Registry backup before a broken entry is touched
  • Exit codes are checked against the live inventory, not trusted alone
02

See where every byte went

A treemap and virtualized tree expose the folders behind a full drive. NTFS MFT scanning is used when elevated; a streaming walk covers the rest.

  • Reparse points are never followed
  • Hard links are deduplicated so allocation is not double-counted
  • Duplicate detection uses size, edge hashes and SHA-256
03

Clean what you actually own

Developer and AI caches are mapped to documented provider paths, then shown for review. Conversation history, credentials and model data stay protected.

  • No arbitrary shell or package-manager execution
  • Nothing is preselected for you
  • Quarantine keeps the run reversible
04

Diagnose the spike, not the vibe

A 20-second capture keeps the peak instead of the latest reading. Kernel time has no process row, so Cyron adds driver versions and recent System events.

  • Kernel share of busy CPU, with the application time it does not explain
  • Display, storage and network driver dates from the package
  • No driver rollback, service disabling or placebo tweaks
Installed applications Review queue
Search apps All sources
  • Adobe Creative CloudRegistered uninstaller Selected
  • Docker DesktopRegistered uninstaller Selected
  • SteamRemoves through Steam Not queued
  • Portable utilityNo registered uninstaller Forced removal
2 selected Review batch
Disk space NTFS · C:

Illustrative treemap. Allocation is measured; no sizes are claimed here.

Known cache locations Review
  • npm cachePackage manager cache Review
  • Cargo targetRust build output Review
  • Hugging Face modelsModel weights Protected
  • Ollama modelsLocal model store Protected
  • Docker / WSL disksVirtual disks Protected
Protected data is never offered Preview cleanup
Real Cyron 20-second spike capture showing peak CPU, kernel and disk values and the busiest process. Real capture · v0.2.0-rc.2

Steps 01–03 are interface illustrations. Step 04 is a real capture.

Built for the agent era

Your agent doesn’t clean up after itself.

AI-assisted building leaves a very specific kind of bloat: caches nobody prunes, build output from a branch you abandoned, and model weights you forgot downloading. Cyron maps the locations it can prove — and refuses to touch the ones that are actually your work.

AI & IDE caches Reviewed catalog
  • Claude CodeCLI cache · debug logs · shell snapshots Review
  • Codex CLICLI logs, recreated on the next run Review
  • CursorCode cache · GPU cache · logs Review
  • Gemini CLISession history + checkpoints Protected
  • Ollama · LM StudioModel weights, measured not offered Protected
  • npm · pnpm · yarn · CargoPackage caches and build output Review
Illustration of the reviewed cache catalog Preview cleanup
Mapped from documented paths
  • Claude Code
  • Codex CLI
  • Cursor
  • Gemini CLI
  • VS Code
  • Antigravity
  • Ollama
  • LM Studio
  • Hugging Face
01 Scan

Known provider paths are measured read-only. No package manager runs, no shell.

02 Review

Each entry shows what it is, why it exists and what happens if it goes.

03 Quarantine

Approved files move to a journaled quarantine you can restore.

04 Rebuild

Your tools repopulate what they need — without the mystery folder you never found.

Measured, not marketed

Fast where it counts.

A directory walk is Cyron's fallback when the MFT is unavailable, and it is still quick. These are the actual medians from one Windows machine, published with the method.

C:\Windows\System32
0.21s

to count 20,557 files and 12.1 GiB

Median of 3 runs · directory walk

Warm OS cache, unelevated, median of 3 runs. These are Cyron-only measurements; no competitor was installed or timed. Bar length is the measured median. Method and environment →

NTFS MFT when elevated and available
Streaming walk automatic fallback, no elevation
5.5–11.1 MiB peak working set in these runs
Read-only scanning never changes the tree

Honest comparison

Different tools. One honest table.

Revo, BCUninstaller, WizTree and BleachBit are good at their jobs. This is a documentation-based capability table, not a speed ranking.

No competitor benchmark exists.

We did not install or time Revo, BCUninstaller, WizTree, TreeSize, BleachBit, Microsoft PC Manager or Sysinternals. The only timed column below is Cyron.

Swipe or scroll the table to see every tool →

Documented capability comparison between Cyron and four other Windows utilities. A documented feature is not proof that it works on this machine, and no competitor timing was measured.

Legend: Yes = documented or implemented, Partial = limited or needs configuration, Reported = third-party report, not verified from the official page, Not verified = not established from the cited source, Not run = no same-machine test. Cyron cells describe this repository and its evidence, not a claim of superiority.

For the late-night spike

When the lag is kernel-deep.

Kernel time has no process row. Cyron shows the kernel share of busy CPU, the application time it does not explain, driver versions and recent System events.

  • A 20-second capture keeps the peak and the busiest process.
  • Drivers are read from the class registry with package dates.
  • A denied event log is reported as unavailable — never as “no problems”.
  • The counters cannot attribute DPC/ISR time to one driver. Cyron says so.
Read the native verification
Cyron driver evidence panel listing AMD, Intel and TP-Link drivers with package dates and recent Windows System events.

Windows 10/11 x64

See what your PC is really doing.

Cyron 0.2.0-rc.2 is a testing candidate. Download it, review the plan, and keep the quarantine until you are sure.

Unsigned testing candidate. Windows may warn and SmartScreen has no reputation for it. This is not a publicly trusted production release.