Release notes

What's new in Langcap

Every release, newest first. Langcap checks for updates in the background and installs them on restart — this is the record of what each one changed.

0.1.1 — Fixes 2026-08-03

A crash worth shipping on its own, a language that was quietly working at half strength, and much smaller downloads.

• Single-channel audio no longer closes the app. A great many podcasts, lecture recordings, and voice memos are recorded in mono, and importing one shut Langcap down mid-import. Fixed, and covered by a test so it stays fixed.
• Korean reads properly now. Korean words change shape constantly, and looking one up means finding the form a dictionary lists it under. Every release until this one fell back to guessing at syllables, so a Korean source produced noticeably fewer usable questions than it should have. The analyser that does it properly now ships.
• Language downloads are about ten times smaller. English drops from roughly 420 MB to 48 MB, and the nine together from 1.7 GB to 161 MB — so the wait when you start a new language is a fraction of what it was, and it takes far less room once it lands.
• Setup no longer shows an internal word in place of "word analysis" while it prepares a language.

0.1.0 — Founding beta 2026-07-20

The first public build of Langcap. The whole idea in one release: drop in the content you already watch, listen to, and read, and study it back as questions stitched to the source — all on your own machine.

• Mine your own media into questions. Drop in video, audio, ebooks, PDFs, or web pages; Langcap cuts them into short questions, each stitched back to the exact clip, line, or page it came from.
• Two modes per source. Play — watch or listen it back with each word lighting up as it's spoken and a translation beneath. Study — the question loop scoped to that one source, asked again until it sticks.
• A library of everything you add, each source its own card, ready to replay or drill.
• Local-first. Transcription (Whisper), question mining, and deciding what to ask next all run on your machine — no account, nothing uploaded, offline after the one-time model download.
• Nine languages. Seven with the full text-or-audio on-ramp (English, Spanish, French, German, Italian, Portuguese, Chinese); Japanese and Korean study from audio today, with their synthesized text voice coming next.
• Windows (needs a Vulkan-capable GPU). macOS later.

It's a young build, and the founding price reflects that — a one-time purchase you own, with the first year of updates included.