Finding files, proving copies, reclaiming storage: FilePerch Windows guide
Nobody loses files because the search was too slow. They lose files at the edges: the copy that silently stopped halfway, the duplicate set that ate a third of the drive, the "cleanup" that deleted the wrong folder. FilePerch is a 100% local Windows file-finder and storage workbench built around a different order of operations — find the file first, verify the bytes before you trust or delete, reclaim with review. It never uploads, never phones home, and never deletes on its own.
Find the file first
FilePerch searches filenames and keywords across your whole drive, then shows matching paths for you to review — no digging through Explorer windows, no "waiting on Windows Search." Its Microsoft Store listing specifies local indexing across 2.8M+ files with sub-15 ms answers, built on a local walk or the Everything index you may already run.
Speed is only half the story; availability is the other. FilePerch is dual-engine by design:
- Everything IPC live resolution — it connects to the Everything index (v1.4, or v1.5 alpha for live IPC) and resolves candidates against the live filesystem.
- Continuous NTFS atomic manifest snapshots — drives without Everything, and moments when Everything is paused or restarting, are served from on-disk manifests, so search never goes blind.
The query modes are deterministic instead of fuzzy: exact-path, basename, and glob queries find the file you named, not a ranked guess at "what you probably meant." Every result is then checked against the live filesystem and marked Verified or Index only, and each result card shows its index freshness state — FilePerch tells you when the index may be stale instead of pretending it is alive. Indexing covers internal SSDs and HDDs, external USB drives, and network shares from one workbench, and metadata indexing runs strictly on the local device.
There is an optional offline semantic ranking step that reranks the index's own candidates with a CPU-only model (BAAI/bge-small-en-v1.5, which requires AVX2 support). It can only reorder candidates the index already supplied, and the Microsoft Store package deliberately ships deterministic lexical ranking — if AVX2 is unavailable, FilePerch uses that deterministic ranking automatically. The search you get never depends on a cloud call.
Where the free tools stop, FilePerch starts
Everything, WizTree, TreeSize, and DocFetcher are genuinely free and genuinely good at one thing each. Category research through JE Horizon's 2026-09-26 comparison notes what happens after the file list appears.
| Tool | What it's good at | Where it stops |
|---|---|---|
| Everything (free) | Filename results in milliseconds | At the file list — no byte verification, no reclaim planning |
| WizTree / TreeSize (free) | Showing what fills the drive | Measure-only — no plan, no proof, no review step |
| DocFetcher (free) | Full-text search inside chosen folders | Per-folder indexing overhead |
| FilePerch | Find, then verify, then reclaim | — |
Finding the file is the first step of the workflow FilePerch was built for. The next two are where things actually get lost.
Prove it: the bytes decide
FilePerch's duplicates feature works by hashing bytes straight from disk, not by comparing names and sizes. Filenames and timestamps lie; SHA-256 hashes don't. The product's own phrasing is the honest one: "not fuzzy name matching — proof of identical bytes." Duplicate groups open read-only, so you see the candidate paths and reason about them before any action is possible.
The same byte-level standard applies to copies and archives:
- SHA-256 integrity verification of copies — every storage-copy job computes and compares the SHA-256 hashes of source and destination from disk. "Did the copy actually copy?" is answered with computed hashes, not a progress bar.
- Durable integrity receipts — completed jobs leave durable receipts, and the product-level exported receipt is fully redacted: evidence you can keep for records and audits without leaking your local paths.
- Pre-flight dry runs and space calculations — copy plans are validated before execution, so you see the whole job, including capacity math, before anything moves.
- Archive-member safety inspection — every zip and tar member is inspected before materialization: member and byte caps, a zip-bomb ratio guard, nesting caps, name collisions, refusal of symlink/hardlink/device/FIFO members, refusal of absolute-path traversal, and destination containment. Opening a hostile archive can't explode into your drive or escape its target folder.
- Archive-before-delete workflow — build the archive, compute its SHA-256, verify the archive's integrity, then confirm deletion of the loose originals. That order is the whole point: you verify before you delete, never the reverse.
One honesty note that belongs next to every hash FilePerch prints: a matching hash proves the bytes matched what was hashed at the times measured. It doesn't declare what the file does, whether it opens in the app you need, or whether you still need it at all. That judgment stays entirely yours.
Reclaim with proof, not with promises
Storage cleanup is where the "just delete it" tools do their damage. FilePerch's safe-reclaim path is a workflow, not a button:
- Storage breakdown — see what is where and what consumes space, with per-drive visibility, before touching anything.
- Review-before-reclaim — inspect cleanup candidates and choose what to remove yourself. FilePerch's honesty line applies here: review before you reclaim — it never deletes on its own.
- Junk-candidate recommendations with confidence levels — stale build caches, temp files, and obsolete artifacts are flagged as candidates using a 9-dimensional adversarial preservation framework, each marked High, Medium, or Low confidence. The recommendations are strictly non-destructive.
- Recycle-bin-only reclaim — safe candidates move only to the Windows Recycle Bin, never hard-deleted. Reclaim is reversible.
- Zero automatic destructive actions — no delete, move, or overwrite happens without explicit interactive confirmation.
Copy and migration plans follow the same philosophy: every source file is preserved in place until you check the completed copy. The worst data-loss mode — "delete the old disk hoping it copied" — is structurally removed because originals are never retired automatically.
Read-only inspection rounds out the workbench: photo-library inspection for RAW/JPEG duplicate triage, hardware discovery to see the sizes and volumes the index covers, and a drives refresh that pulls real volumes into one view. Inspect first, decide later.
Browse without breaking things
Search results can open in the Windows default viewer, reveal their location in Explorer, or go to the app's settings vault — one click, no hunting. Because search results can come from folders you haven't looked at in years, opening an executable or script result requires explicit confirmation first. Protected-path filters and policy guards also prevent accidental execution of binaries and exposure of credential stores while you inspect folders: browsing an unknown archive can't quietly run something or hand out a saved password.
The app's own trial state, license state, and settings persist in a Windows current-user DPAPI-protected vault — machine-local, user-scoped settings protection, kept separate from the files you work with.
Ask in plain English
You can ask questions in natural language — "find photo duplicates with these keywords," "run a safe reclaim" — and get executable results back as interactive hitcards. The design constraint that makes this trustworthy instead of spooky: deterministic code owns every path and action. The assistant can only reorder candidates the index already supplied; it is architecturally unable to invent a path. The pricing page states it as bluntly as possible: "No hallucinated file paths."
Directory autocomplete is powered by live NTFS trees rather than a stale cache, errors come back in plain English with a cause, a recovery step, and the technical detail underneath, and double-clicking a result opens the file in its default application while Show in Explorer locates it — with file-existence errors surfaced honestly when the file is gone.
The privacy contract
Search, metadata indexing, semantic ranking, and storage analytics all run on the local device: zero cloud inference, zero external API keys, zero telemetry, zero network traffic. "Your file inventory never becomes a product" isn't a slogan here — there are no endpoints to send it to. For records and compliance staff who cannot upload proprietary folders to third-party clouds, that is the purchase driver rather than a footnote.
The license honors the same boundary. Subscription licenses are validated 100% offline against machine identity with cryptographic Ed25519 signatures, bound to a privacy-preserving device code — FilePerch doesn't phone home even to check that you paid. The pricing page is explicit: "No background phoning home — and no lifetime purchase." On expiry, the app blocks new paid work (startup, timers, pre-dispatch, queued jobs) while existing results, open views, and in-flight copies remain available; importing a renewal restores normal operation.
One packaging detail spelled out for the record: packaged releases are secured with AES-256-GCM-sealed proprietary modules at rest, loaded into memory at runtime. That is packaging protection for the app's own code — it is not a claim that your files are encrypted. FilePerch verifies your files; it doesn't encrypt them.
Platform and pricing
FilePerch is a native Windows 10/11 x64 desktop application (PySide6) — no web views, no cloud mocks, just local data rendered on your own hardware. It's distributed through the Microsoft Store, with requirements disclosed up front: Everything 1.4 or 1.5 alpha for live IPC indexing (manifest snapshots cover drives without it), and AVX2 for the optional semantic ranking. The Store listing governs current pricing, system requirements, and updates.
Pricing is a subscription, stated plainly:
- $29.99/yr or $3.99/mo after the trial — FilePerch · 1-Year Subscription and Monthly Subscription, exactly as published.
- Free download; free 30-day automatic trial — the trial is on the Microsoft Store page, no card required to start.
- No lifetime purchase — subscription only, per the pricing page.
- Every purchase includes the Wittlet Desktop Companion free ($12 value) — Wittlet is JE Horizon's offline AI companion for Windows.
Compared with the category, the price sits low: JE Horizon's 2026-09-26 research priced Copernic Desktop Search's entry tier around $28.99–$95.99 per user per year and Acronis True Image at $49.99–$124.99 per year, while Everything, WizTree, TreeSize, and DocFetcher are free but stop at search or measurement. FilePerch undercuts the paid tools and out-integrates the free ones at $29.99/yr or $3.99/mo after a free 30-day trial.
The pattern that holds the whole product together is the same one that makes it worth paying for: it doesn't say the file is there, the copy is good, or the cleanup is safe. It shows you the bytes, and it only acts after you review. Find the file. Prove the copy. Reclaim with proof.