Find what’s twice. Keep what matters.
Find duplicate files and similar photos
Duplicates proven byte-for-byte. Similar photos ranked and explained by AI that runs on your machine — no account, no upload. And nothing is deleted without you. New in 2.1: your Photos library, on a Mac.
Free shows you the results, not a paywall. The cap is on how much you see, never on how well.
Every download starts with 7 days of Pro.
Two engines, one promise
Which half is yours?
Duplicates are proven. Similar photos are ranked and explained. They never pretend to be the same thing.
Duplicate files
Any file type, any drive, including NAS.
- Full BLAKE3, byte-for-byte — zero false positives
- Never guesses: if it says identical, it read every byte
- APFS clones flagged, so you don’t “free” space that isn’t there macOS
Free: 25 files per clean · Standard: unlimited
Similar photos Pro
Bursts, exports, re-saves, near-identical frames.
- Grouped by what they look like, not by name or date
- You set how tight a match counts, and nothing leaves a group until you say so
- Review with AI ranks each group and says why: sharpness, exposure, eyes open
Free & Standard: first 8 groups, whole · Pro: every group
The scan
You choose what it
reads.
Point it at folders — or, on a Mac, at your Photos library. Exclude the ones you want left alone. Set a size floor. Then it runs.
Skip system files, hidden files and folders, library containers, photo libraries, NAS and system folders — or name your own
Include or exclude by extension, and set a size range anywhere from 1 KiB to unlimited
Follow symlinks or collapse them, so a link never counts as a copy of the file it points at
All of it under Customize…, with defaults that are already sensible. The photos side adds two more: how close a match has to be, and how large a group may get.
Fast on real data, not just benchmarks
Up to ~20× on local files, ~1.2–1.4× on NAS volumes
Real runs against Czkawka, dupeGuru, and Gemini 2 on code, design, and NAS datasets. DuoBolt finishes jobs where other tools slow down or time out.
Measured on DuoBolt 1.4, M1 Pro, caches cleared between runs — pending a re-run on 2.0. Gemini 2 was run in duplicates-only mode (similar detection disabled). dupeGuru and Gemini 2 timed out on NAS; Gemini 2 couldn't scan Library.
Why these results hold up
Each dataset was scanned back-to-back on the same machine with identical include/exclude rules. DuoBolt's engine keeps all CPU cores busy with two-stage BLAKE3 hashing; competitors process files more sequentially.
- Per-root parallelism and BLAKE3 worker hashing keep multiple CPU cores busy instead of going file by file.
- Streaming, chunked reads overlap disk/NAS I/O with hashing, hiding latency on slower disks or remote drives.
- The interface runs on the same Rust engine that does the hashing, so responsiveness never costs byte-perfect accuracy.
- Chunked scan stages overlap hashing and comparison so CPUs never sit idle waiting for I/O, keeping throughput high even on NAS.
- The run
- Machine
- Apple M1 Pro, macOS
- Datasets
- 5, up to 1+ TB
- Method
- Back-to-back, caches cleared
- Version
- DuoBolt 1.4 — re-run pending on 2.0
Somebody else’s measurement
“It’s very fast, surprisingly so.”
Every figure in the section above is ours, measured on our own machine. This one is not: an editor at Softpedia installed DuoBolt 2.0.2, pointed it at his own folders, and rated it 4 out of 5.
The same review calls DuoBolt “quite expensive for a
duplicate cleaner”, and it tested ordinary folders on a Mac
— no NAS, no RAW, no ~/Library.
- Source
- Softpedia
- Reviewer
- Catalin Chelariu
- Version tested
- DuoBolt 2.0.2
- Date
- August 2026
AI Curation
It marks the best shot in
every group.
Open a group and DuoBolt says which of the others are only copies of it — and which ones are different enough that you should look yourself.
The groups it is unsure about go to Needs your eye, and the ones with nothing to do say so and get out of your way.
Explainable, never automatic
It ranks. You decide.
DuoBolt never deletes a photo on its own. It works out how two shots are related before it scores them, measures only what applies to that relationship, and shows its working — so you can disagree with it in one click.
When the evidence is thin it says so instead of inventing a winner. Groups it is unsure about go to Needs your eye rather than into a selection.
A copy of the kept shot. Same scene, written again — nothing the kept shot lacks.
- Face sharpness
- 3.5× sharper
- Face exposure
- Worse exposed
- Resolution (MP)
- Same, 19.6 MP
Compared with Girl_with_a_Pearl_Earring.jpg
Formats
It reads what your camera writes.
Similar-photo grouping works on RAW files directly — not just on the JPEGs you exported from them. So the originals get culled too, before they eat another drive.
Finding duplicates never decodes a file. It hashes bytes, so a duplicate HEIC turns up on Windows exactly as it does on a Mac. Grouping similar photos does have to read the picture. RAW works on both.
1 On Windows, HEIC and AVIF need two free Microsoft Store add-ons: HEIF Image Extension and AV1 Video Extension. Without them those files are skipped silently. macOS needs nothing.
Safety
The same promise on
both sides.
Proven duplicates and ranked similars are different jobs, handled the same way.
Nothing moves until you say so. Every group is laid out with paths, sizes and previews first. DuoBolt never acts on a scan by itself — not on duplicates, not on photos.
Four ways out, and you pick. Move to Trash, move to an Archive folder, copy somewhere else, or delete outright. Only the last one is final, and it says so. A photo from the Photos library goes to Photos' own Recently Deleted.
Everything is written down. Scan history and operations history record what ran and what it touched, so you can restore a file later — or just check what happened last month.
Folders you can’t touch by accident. Mark any folder as protected and its files stop being selectable. The checkbox refuses to tick, rather than warning you afterwards.
On-device AI
Your photos never
leave your machine.
Photo AI usually arrives with an account attached, and often with your library on somebody's servers. DuoBolt ships the models instead — no sign-up, no queue, no upload. It works on a plane.
No account, ever. Nothing to sign up for, nothing to log into, nothing to keep paying for.
No upload. Inference runs in-process. The only network calls DuoBolt makes are license checks and the update check.
Telemetry off at startup. And the macOS runtime we ship contains no telemetry code at all — verified by a byte scan on every update.
Duplicate finders and cleaners
Account required · per machine, per year
per year
DuoBolt Pro
Two machines · no account · runs offline
once
Category range for subscription duplicate finders and Mac cleaners. Yours forever, including every 2.x update. Both tiers, in full →
Three different messes, one workflow
A photographer deciding between near-identical frames, a developer with dozens of cloned repos, a NAS filling up with the same films twice. The engine differs; the review-before-anything-moves part does not.
Photographers
A wedding leaves you eight hundred frames and forty of them are the same moment. DuoBolt groups them, ranks each group on sharpness, exposure and whether the eyes are open, and tells you why it chose. On the RAW files, not just the exports — and on a Mac, inside the Photos library itself, imports and favorites included.
How culling worksDevelopers
Cloned repos, node_modules copies, archived build artifacts, and forgotten project backups accumulate fast. DuoBolt scans a 38 GB codebase in 3 seconds, groups what is byte-identical, and lets you review before a single file moves.
How exact matching worksNAS & home server users
Media libraries on Synology, QNAP, or any network drive grow over time with duplicate movies, photos, and music from multiple devices. DuoBolt scans a full terabyte of NAS data in under 90 seconds even on cold, spun-down disks.
See the NAS guide