Release notes

Latest: version 1.3, October 2026

What changed in each version of Hefty Search, and what it means when you use it.

Version 1.3

A query is read for what it asks, not how it asks. "Find the photo of the beach at night" looks for the beach at night, in photos.
Spoken lines are matched as a phrase, word for word and in order. A quote finds the moment it was said, not every line that shares a word.
Results are no longer cut off at a fixed number. Each kind of file shows everything that clears its bar, and weak picture matches are left out.
Slash commands narrow a search from the bar: by kind of file, by folder, by date, or newest and oldest first. Type /help to see them all.
Your words are encoded in 8 milliseconds, down from 42, and a library of 63,000 items is ranked in about 15 on an M1 Pro. Results keep up with your typing.

Picture results now stop where they stop being right. In a check of 200 searches, the last picture shown was the right one 97% of the time.
A picture search uses your words as you wrote them. The stock phrase once added to them, which let one group of lookalike photos match anything, is gone.
Search no longer swaps in synonyms. A search for food stops showing plates because of the word dish.
Matches by meaning are judged against your own library. Loosely related documents and spoken lines stay out of the results.

Indexing has two speeds in Settings, Slow and Fast. Slow stays out of your way; Fast runs the Neural Engine and the GPU together.
Fast indexes about 36 photos a second on an M1 Pro with 16 GB, three times the standard pace.
Macs with more GPU cores and memory run more copies of the photo model side by side, and index faster still.
Speech is transcribed about 15 times faster than it plays. An hour of talk takes about four minutes.
The indexing panel shows photos per second and time elapsed, so you can see how a run is going.

A video is now read in one pass: scene cuts pick its key moments, speech is transcribed, and each moment is described. Find it by what is shown or said.
Text on screen is part of each moment's description. A slide, a sign or a caption in a video can be searched like anything else.
Speech is found by a voice detector, then transcribed by Whisper small with a decoder of our own. Quotes land on the right second, and a badly heard passage gets a second try.
A video is written to the index all at once when it is done. Quitting halfway never leaves half a video behind.

Each document gets a title and search terms written by a language model on your computer. It turns up for its subject, even when your words are not in it.
Photos are indexed in a single pass, with no separate text reading step. Indexing is quicker, and words on a photo are still found.
Re-Scan Library now re-indexes in place. Search keeps working while it runs.

The index is encrypted at rest with AES-256, the key kept in your login Keychain. Transcripts, face data and search vectors cannot be read outside the app.
Thumbnails are no longer stored. Result cards are drawn from the file itself, which gives the disk space back.

Models are prepared in a helper that survives a quit, and a damaged model cache rebuilds itself. Closing the app no longer throws that work away.
If a model cannot load, the app says so and recovers. The search bar no longer stays stuck.
The license check runs in the background. A slow network can no longer make the app hang at launch.
A forced re-scan no longer lists the same spoken line twice.
Moved files are matched more carefully. Two files that only look alike are no longer taken for one file that moved.
Face scanning stops cleanly. Removing a person, deleting face data or turning the feature off ends the scan first.

Software Update has its own place in Settings, under Advanced: automatic checks, automatic installs, a manual check and the version you are running.
The app's data folder is now named Hefty. Your index moves across on first launch, with nothing to redo; that launch takes a little longer, once.

Search with a screenshot. Capture any part of the screen and find the pictures that look like it.
Video search by scene. A result opens the video at the moment that matched.
Rename a file from its result card. The index follows the new name.
Auto-sync keeps indexed folders up to date as files change. New files are searchable without a re-scan.
Choose which kinds of files are indexed, and skip a slow file in the middle of a run.
Remove one folder from the index without touching the folders inside it that you added separately.
Launch at Login, and a bug report form in Settings that reaches us by email.

Questions? Email [email protected].