Regex+ comes to JavaScript and TypeScript. Steven Levithan's superset of JavaScript regex gives you atomic groups, possessive quantifiers, subroutine calls, and the n (named-captures-only) and x (free-spacing) flags. It compiles down to standard JavaScript, so it runs on the engine already shipping — and it is free on every tier, including the free one. The rest of this release is correctness. Every flavor is now checked construct by construct against its real reference engine, and the disagreements that remain are written down with a named cause rather than left to be rediscovered. Shell is three tools instead of one. Choose BSD grep, sed or awk and the engine behaves like that tool: its dialect, its quirks, and which backslash escapes it silently ignores rather than honours. R runs both of its engines in-process — TRE, which base R uses by default, and PCRE2, which perl = TRUE selects. R's own POSIX character-class tables are ported, so [[:punct:]] answers exactly as R does. No system R needed. When a pattern will not compile, the app now tells you what is wrong and what to change, in all four interface languages, instead of forwarding the engine's own English. Each message is written for the engine that produced it: Perl's lookbehind limit, Java's and PCRE2's are three different rules and now read as three.

The VS Code extension is published. It detects regex literals across 20 languages and keeps a live two-way session with the app: edit the pattern in your editor and the canvas follows, rework it on the canvas and the change is written back to your file. Reconnection no longer gives up, so launching the two apps in either order works. Replace was broken on several engines and is fixed. The Shell engine dropped every capture group. Dart's replace returned the wrong result for every input. PostgreSQL and MySQL were being sent JavaScript replacement syntax, so backreferences silently did nothing. Match positions are correct across the board: five engines reported offsets in the wrong unit, which put highlights in the wrong place on any text containing non-ASCII characters. Java's case-insensitive flag no longer quietly turns on Unicode case folding, and \w now matches what Java and R actually match rather than what PCRE2 does by default. The canvas tells the truth about more constructs: \h, \v and \N are modelled as the escapes they are instead of plain letters, negated escapes and set operations are labelled correctly, and characters outside the basic plane are reported as whole code points rather than surrogate halves. For Rust, you now choose the engine — the linear-time regex crate or fancy-regex — and that choice is honoured everywhere, including generated code. Also: clearing the workspace is undoable, switching flavors no longer strips flags your saved patterns rely on, and AI requests are made for the flavor on the canvas instead of always JavaScript.

JavaScript and TypeScript now run on JavaScriptCore, the regex engine built into macOS and used by Safari, instead of a Rust reimplementation of the spec. Patterns behave exactly as they do in a browser, and ES2024 unicodeSets mode is supported: the v flag now works, so set notation like [\w&&[a-z]] and [\p{L}--[a-z]] does too. Fixed: the Dart, Elixir and R engines could quietly fall back and report no matches at all, which looked identical to a pattern that genuinely did not match. They now return real results, or a real error. Match accuracy fixes across several engines: PostgreSQL zero-length matches, R control characters and CRLF handling, Elixir Unicode and NUL bytes. Character class editing: dropping something that cannot live inside a character class now places it beside the class, instead of quietly corrupting the pattern.

Maintenance release. No changes to app features or behaviour; this update refreshes packaging and the App Store listing.

bug fix that solved accidental clearing of input text when editing the regular expression manually