What's new

EasyHomebrew 3.7.0 brings taps back. Since June, Homebrew ignores everything from a third-party tap until you tell it you trust it, so packages from taps quietly stopped working. The Taps page now shows how far Homebrew trusts each tap and what you installed from it, and trusts the packages you use, or a whole tap, in one click. You add a tap by its name or its repository's address, including taps that aren't on GitHub, or pick one of the taps people install from most, and choose how far to trust it. Packages from taps appear in All Packages and search with a TAP badge, removing a tap also removes what you installed from it and the trust you gave it, and Installed, Updates and Doctor tell you when Homebrew is ignoring something you installed. Backups keep your tap packages, and updates to packages from taps show up again. The update is free and the app offers it itself; the Setapp edition follows once Setapp has reviewed it.

• Taps work again: see and set how far Homebrew trusts each tap, in one click
• Add a tap by its name or its repository's address, even off GitHub, or pick a popular one
• Packages from taps in All Packages and search, with a TAP badge
• Removing a tap also removes what you installed from it and the trust you gave it
• Installed, Updates and Doctor tell you when Homebrew is ignoring a package

EasyHomebrew 3.6.0 is a security release. Vulnerability checking is stronger: far more of your installed packages are now checked against public security advisories, each issue shows how serious it is and the version that fixes it, and the packages no advisory database follows are listed as such instead of being counted as safe. You can update just the vulnerable packages, from the app or automatically from the menu-bar agent, and every package has a Security tab of its own. Above all that, a new Integrity page makes sure nothing has tampered with what Homebrew installed: the files inside your packages, Homebrew itself and its taps, and the signatures of the apps it installed, with Homebrew's own copy put back in one click. Optional safeguards make Homebrew check every download before it is installed, and EasyHomebrew applies them every time it runs Homebrew. Every action now shows its progress in the middle of the window and waits its turn when Homebrew is busy, Doctor clears up deprecated packages that others still use, and Updates can be shown as a grid. It all runs on your Mac. The update is free and the app offers it itself; the Setapp edition follows once Setapp has reviewed it.

• Stronger vulnerability checking, with the severity of each issue and the version that fixes it
• Integrity checks: nothing has changed your packages, Homebrew itself or the apps it installed
• Optional safeguards: Homebrew checks every download before it is installed
• Security updates only, by hand or automatically, and a Security tab for every package
• Every action shows its progress and waits its turn when Homebrew is busy

WizJS has a map that keeps its keys in order. BBTreeMap holds string or number keys sorted, so you can walk them in order, take every entry between two keys, and ask for the first or the last, which JavaScript's own Map cannot do. It follows bbtree, Xiangpeng Hao's Rust B-tree, which calls itself the fastest ordered map, and we measured which of its claims still hold in JavaScript. Most do. At ten million entries its lookups are 2.75 times as fast as a classic B-tree's for numbers and 2.35 times for UUIDs, about what the Rust original reports; it is 14 to 18 times as fast as a skip list, takes keys in ascending order at about 20 nanoseconds each, and needs less than half the memory for number keys. Two claims do not carry over: JavaScript has no way to fetch memory ahead, so asking for many keys at once gains nothing, and string keys cost more memory than a plain B-tree's. When you never need the order, a Map is still faster. The classic B-tree comes too, as BTreeMap, with a comparator for keys of any type. The whole benchmark, claim by claim, is in the repository, and you can run it yourself.

• BBTreeMap: an ordered map for string and number keys, with range, first and last
• Lookups 2.75× as fast as a classic B-tree for numbers, 2.35× for UUIDs, at ten million entries
• 14–18× as fast as a skip list, and less than half the memory for number keys
• BTreeMap, the classic B-tree, takes a comparator for keys of any type
• Every claim measured, with the benchmark in the repository

Showcase is on the Mac, as a beta. Showcase for Mac 1.0.0 is the Windows app written again for macOS: the games you own on Steam, GOG, Epic Games, PlayStation and Nintendo in one library, copies bought twice stacked into one poster, a profile page for every game with its achievements and screenshots, the stats, and the three brag cards, your numbers, your rarest achievements and what your collection costs, saved as an image or copied. Change poster, full-size posters from SteamGridDB with your own key, the cartridges and discs you enter by hand and the Hidden, Installed, New and No art filters are all there. The library is the same file as on Windows, so a library.db copied from a PC opens on the Mac and the other way round. You sign in on each store's own page, and the sign-ins live in your login keychain; the app is sandboxed, and asks once for a launcher's folder before the Installed filter can read it. It is free, for macOS 14 or newer, one notarized download for Apple silicon and Intel Macs, and it updates itself. It is a beta: newer, and less tested against the live stores than the Windows app. The Mac App Store edition follows.

• A beta: the Windows app written for the Mac, with five stores, stacks, profiles, stats and the three brag cards
• The same library file as on Windows: copy library.db between the two and it opens in both
• Store sign-ins in your login keychain; sandboxed; the Installed filter asks once for each launcher's folder
• Free, macOS 14 or newer, Apple silicon and Intel in one notarized download that updates itself
• The Mac App Store edition follows

Whokoban's next update trades its chiptunes for a band: accordion, trumpet, clarinet, tuba, banjo and glockenspiel, with a bass drum, a snare, a woodblock and a tambourine. It has four tunes for the levels and a polka for the title screen; each of the six seasons brings the polka in its own dress and four tunes more; and Herbie gets a tune of his own in the FlyCraft look — thirty-six in all. None of it is recorded. The game plays every note as it goes, and so does the new soundtrack page, in your browser, so you can hear every tune before 2.5 reaches the App Store. A slider there stands in for the move cap: in the game, the closer you get to it, the faster the band plays and a clock starts ticking. In 2.5, the speaker on the title screen opens a listening booth with every tune, in season or out.

• A band in place of the chiptunes: accordion, trumpet, clarinet, tuba, banjo and glockenspiel, with drums, a woodblock and a tambourine
• Thirty-six tunes: the band's five, five for each of the six seasons, and Herbie's Flight for the FlyCraft look
• Every tune on the soundtrack page now, played the way the game plays it
• A slider for the move cap: the closer you get, the faster the band plays
• In the game, a listening booth behind the speaker on the title screen
• Coming in Whokoban 2.5, a free update; the App Store has 2.4.0 today

EasyHomebrew 3.5.0 is about seeing more at a glance and finding the right tools. There is a category for AI now, for chat apps, coding agents, local models and image generators. Toolkits are a new kind of persona for people who know the job but not the tools: twelve of them gather the best open-source apps for web browsing, local AI, photo editing, video, design, music, 3D printing, office work and privacy, each installed in one step, and an AI Coding persona brings Claude Code and Codex. The sidebar now says how things stand without opening a page: a green check when nothing you have installed has a known vulnerability, a sign on Doctor, a warning when a backup is overdue, how many services are running, and how many new packages each category and persona has. Personas show which of their packages have an update and update them together, the menu-bar agent's menu is simpler, with its settings moved into the app, and package icons stay loaded. The update is free and the app offers it itself; the Setapp edition follows once Setapp has reviewed it.

• An AI category for chat apps, coding agents, local models and image generators
• Toolkits: the best open-source apps for a job, from photo editing and video to local AI, each in one step
• An AI Coding persona with Claude Code and Codex, and AI Engineer tools for running and fine-tuning models
• The sidebar at a glance: Security, Doctor, Backup and Services, and new packages counted in every category
• Personas show which of their packages have an update, and update them together
• A simpler menu-bar agent, set up from the app

Showcase 1.3 adds two brag cards and puts the posters in your hands. In Stats, under the brag card, a new card adds up what your collection costs: Read prices asks Steam, GOG, Epic, PlayStation and Nintendo what every game you own sells for today in your country, in its currency, and the card gives the total at the regular prices, what it would cost with today's sales, each store's share and your three priciest games. A second card counts the achievements you have unlocked that fewer than 1 in 10 players have, and shows six of them with their game and how rare each one is: your rarest, until you choose the six you would rather show. Both copy or save as an image, like the first. Right-click a game and choose Change poster to pick a better one: one of your copies' own, a store's cover, SteamGridDB's community art, or an image file of your own. Your pick shows everywhere in Showcase, and syncs never undo it. PlayStation and Nintendo art is square and Epic's is wider than a poster, so give Showcase a free SteamGridDB key and every game without a full-size poster gets one on its own, only from the exact same game; SteamGridDB is the one service besides the stores and GitHub that Showcase talks to, and only with your key. The filters above the grid are icons now, and a new No art filter finds the games still showing a Showcase cover. Nintendo syncs again after Nintendo changed its purchase history page, Epic's covers are shown whole, and Steam achievements that share an icon no longer all show as unlocked when you have one. When an update is out, Showcase shows its notes in a window of its own, with Install and restart beside them. 1.3.1 followed the same day: in a country with no English PlayStation Store, PlayStation games are priced from the country's own store instead of in US dollars, and Change poster no longer offers soundtracks, DLC or games unrelated to the search. Installed copies are offered the update by the app itself.

• What your collection costs: today's prices from five stores, in your currency, on a brag card
• Your rarest achievements: the ones fewer than 1 in 10 players have, six of them on a card
• Change poster: your copies' own, a store's, SteamGridDB's or a file of your own, kept through every sync
• Full-size posters from SteamGridDB for games with square or wide art, with your free key
• Filters as icons above the grid, and a No art filter for the games still on a Showcase cover
• Nintendo syncs again, PlayStation prices from your own country's store; still free, for Windows 10 or 11 (64-bit), as a .msi installer

This release started as a list: every place where EasyHomebrew, used on a real Mac, left someone asking what a thing meant. 3.4.0 works through it. The menu-bar agent could only be started or stopped from Settings, three clicks away from where anyone notices that it is not running; it now has a button in the bottom-left corner of the window, with a dot that says how it is doing, and nothing joins your login items unless you press Install. The health score was a heart and a number (nobody could tell what 80 was out of); it is now a word, Good, Fair or Poor, and clicking it shows each thing that costs points and where to fix it. Doctor runs when the app opens, so its warnings count from the start. Updates is one list, and a stale dependency sits inside the package that needs it instead of in a section of its own that explained nothing. Backups are kept, with their dates, ready to restore here or to copy to a new Mac, and once a day the app makes one by itself if what you have installed has changed. About a hundred labels that stayed in English in every language are translated. The update is free and the app offers it itself; the Setapp edition follows with 3.5.0.

• The menu-bar agent is started, stopped and installed from a button in the corner of the window; nothing joins your login items unless you press Install
• The health score is a word (Good, Fair or Poor), and clicking it shows what each problem costs
• Doctor runs when the app opens, and its page leads with a verdict
• Updates is one list; a stale dependency sits inside the package that needs it
• Backups are kept with their dates, and a daily automatic backup follows your changes
• About a hundred labels that stayed in English are translated into all six languages

Kyori's last rule was its oldest, and it was wrong about one thing. When a completion carried the words you typed somewhere other than the start, it charged for the length difference and for how far each word had moved, so 'pizza' against 'margherita pizza' cost 22, more than an unrelated term of the right length. On a filtered list that never showed, because every candidate already began with what you typed. Ranking a whole catalogue, which is what Kyori Index does, it put rephrased completions behind terms that merely began the same way. 0.7.0 rewrites that rule around the words you typed: for each one, the fewest edits to the start of any word in the completion. A completion that has your words costs little ('margherita pizza' is now 2), one that lacks a word costs more than any typo can, and all of it sits below every completion that extends what you typed, so the order you had on a filtered list is untouched; the completion's length no longer counts at all. Measured on the held-out Amazon searches of the progress report: on filtered lists nothing moves (0.404 before and after). Over the whole catalogue the right suggestion moves up from 0.567 to 0.597 in mean reciprocal rank, ahead of Fuse.js at 0.579 where it used to trail. Where the prefix filter had lost the answer to a typo, from 0.365 to 0.583, now 0.03 behind character trigrams instead of 0.25. On 400 real misspellings, from 0.863 to 0.882, ahead of Damerau-Levenshtein. Each result was confirmed once on the test split. This is a breaking change even though nothing is renamed: distance() returns a different number for every pair the last rule decides, and rank(), rankKey() and Kyori Index search() order those pairs differently. similarity() and the other four methods are untouched. The upgrade notes in the package show which pairs moved.

• The last rule now reads the words you typed: a completion that carries them elsewhere costs little
• 'pizza' against 'margherita pizza' costs 2 where it cost 22
• Filtered lists are unchanged; over a whole catalogue the right suggestion moves up from 0.567 to 0.597, ahead of Fuse.js
• Where a typo had lost the answer, from 0.365 to 0.583; on real misspellings, from 0.863 to 0.882
• Breaking: distance() changes for every pair the last rule decides, and the orders built on it; similarity() and the other methods do not

Kyori 0.6.0 does the same work in less than half the time. Ordering an autocomplete list used to spend three quarters of its time tidying the same strings again for every comparison: stripping accents, lowering case, reading dashes as spaces. It now tidies each string once, and a query over a filtered list takes 52 microseconds at the median instead of 120, on the same held-out Amazon searches the progress report measures. Nothing you get back changes. Every distance, similarity, order and index search is identical to 0.5.1, checked pair by pair over a hundred thousand pairs and three thousand lists, and the research tables stand as they were; only the timings were measured again. Upgrade and keep your tests.

• A query over a filtered list in 52 µs at the median, down from 120
• Every result identical to 0.5.1: distances, similarities, orders and index searches
• The research tables are unchanged; only the timings were measured again