What we shipped

Releases with real user-visible changes. The same stories the app tells in its What's New notes; internal plumbing does not make this list.

iOS 1.9.0

September 2026

Japanese, Chinese, Korean and Hindi. Pick your language in Settings or during setup and VAYS selects the multilingual speech model for you (NVIDIA Nemotron, about 650 MB, fully offline). Speaker separation works in these languages too.
Recording in the car, finished properly. Building on the 1.8.5 fix, recording now also rides out longer microphone takeovers, like navigation prompts or longer calls, and resumes on its own.
A new speech engine: 28% faster on our benchmark, and no more stray characters in short Polish phrases.
Lighter speaker detection. The new speaker model is less than half the size and much easier on memory; it downloads fresh the first time you use it.
Settings tell the truth: a new Active engine row shows which model actually transcribes your language, and a note with no recognized speech now says so plainly. After updating you see a short What's New, once.

September 2026

Japanese, Chinese, Korean and Hindi. Pick your language in Settings → Transcription and VAYS selects the multilingual speech model for you (NVIDIA Nemotron, fully offline). Speaker separation in sessions works in these languages too.
A new speech engine: more accurate at the seams of long recordings, 28% faster on our benchmark, and no more stray characters in short phrases.
Transcription settings now tell the truth: you can see which engine actually runs, and the multilingual model has an explicit download and delete button. Picking a language no longer requires Advanced mode.
Refreshed speaker models with more consistent clustering; they download fresh the first time you use them.

August 2026

Better behavior on older iPhones. On devices with less memory VAYS now guards the AI features and defaults to clean transcription, so a recording no longer dies mid-processing.
Transcribe is the default recording mode for every new install; Memo with AI analysis is one swipe away.
A finished transcript now survives even when iOS kills the app during AI analysis; you keep the text and can retry the summary.
New Device check in Settings: a speed test of the whole pipeline on your phone, plus clearer diagnostics when something fails.

August 2026

A notification no longer ends your recording. A message or call at the wrong moment could stop a recording after about a minute, most often while driving with CarPlay. VAYS now returns to recording on its own.
Apple Watch stops pretending to record. If the system denies microphone access you see a clear message instead of a timer counting into the void.
The Activity Log now records processing stages: start, model loading, transcription end. If the app is killed mid-way, the log tells you at which stage.
Fixed the Discord support invite link.

August 2026

New Library tab: browse recordings by project, not just by date. Projects form a tree: type "Podcast/Season 3" at import and VAYS builds the structure for you.
Renaming a project fixes it across every recording underneath at once.
AI analysis is now a switch at import. Bringing in a backlog and you just want the text? Turn it off and skip summaries you never asked for.
Pick Ultra fast or Best quality per import batch, and a separate analysis template for imports.

Older releases live in the App Store version history. See how we measure what we ship.