Kōrero v1.43.0 - see what's in use, a local model that fits, and long meetings read in full

Kia ora! 1.43 makes it easy to see what Kōrero is using, to run AI clean-up on your own computer with a model that fits it, and it fixes long meetings getting notes from only part of the transcript.

New in 1.43.0

• Always see what's in use. The foot of the sidebar shows your microphone (the real device, even on "system default"), the speech model and the clean-up model with its prompt, on every page. Click any of them to change it. The clean-up line says where text goes: on this computer, or the service it's sent to. The microphone and speech model can't be changed mid-recording, and the menu says why.
• A local AI model that fits your computer. AI clean-up & notes now opens with a card that reads your graphics card, memory and free disk, suggests one model that fits, and downloads it with progress and Cancel, test-runs it and switches to it. It installs or starts Ollama if needed. Other models are labelled Fits well, Fits, Slower or Won't fit. The suggestions haven't yet been tested on Kōrero's clean-up prompts, so treat them as a starting point.

Fixes

• Long meetings no longer get notes from half the transcript. Ollama reads only 4,000 tokens by default on graphics cards under 24 GB, so notes and Ask for meetings longer than about 15–20 minutes were written from part of the meeting, without saying so. Kōrero now asks Ollama for a context sized to each request, refuses rather than cuts a transcript that can't fit, and keeps the model loaded at that size.
• Meeting notes on a local model are much faster straight after a meeting. Kōrero's speech model and the AI model were sharing the graphics card, which made reading a long transcript about 20 times slower on a 12 GB laptop. Kōrero now frees the speech model's memory before a local model reads a long meeting (unless you've set the speech model to stay loaded). Your next dictation reloads it, which takes a few seconds.
• Time limits for local models now allow for slower computers, so long notes finish instead of timing out.
• The Today panel names your microphone instead of saying "Default".

Installing

Download Korero_1.43.0_x64-setup.exe below, or let Kōrero update itself. The installer isn't code-signed yet, so Windows SmartScreen may warn you: choose More info → Run anyway. Your settings, meetings, notes and recordings are kept.

Ngā mihi!

Kōrero v1.42.0 — Today, Ctrl K search, and a new Meetings page

Kia ora! 1.42 is the biggest change to how Kōrero looks and works since the fork: one consistent design, a home page that tells you what's happening, search across everything, and meetings you can run from a small window beside your call.

New in 1.42.0

• Today. The new home page. Start a meeting, a note or an import, see what's running and what just finished, and check that Kōrero is ready (microphone, New Zealand English, speech model, notes model and call audio), with a fix button for anything that isn't.
• Search everything with Ctrl K. Meetings (titles, transcripts and notes), notes, dictations and settings in one box. Macrons are optional: "whanau" finds "whānau".
• A new Meetings page. Your meetings sit in a list beside the one you're reading, with Transcript, Notes, Audio and Ask tabs. While you record, the live view shows both sides' levels and the transcript as it arrives.
• Flag a moment. Press F (or Flag moment) during a meeting, and the notes give each flagged moment its own line.
• Pop-out recorder. A small window that stays on top of your call, with the clock, Pause, Flag and Stop. Recording now carries on whichever page you're on.
• A short wrap-up after Stop says what was saved, how much was heard from each side, and anything to check.
• The dictation pill says more. It shows NZ when New Zealand English is on, a clock and a real Stop button in hands-free mode, and the prompt it's cleaning up with. If a paste fails, your text goes to the clipboard and the pill tells you. If there's no microphone, the pill says so, and its Fix button opens the right setting.
• Activity. One panel for what's running, what needs a look, and what finished.
• Notes keep your dictation. A dictation always lands in the note it started in, even if you change page, and Cancel now works.
• Clean-up with local thinking models works. Ollama models such as gemma4 think by default and ran past the time limit. Kōrero now turns thinking off, so clean-up and meeting notes finish.

Fixes

• "Check for updates" in the tray works again.
• Moving recordings to a new folder can no longer save an empty meetings list when your meetings failed to load.
• Dates read the New Zealand way, and settings use plain-English names.
• Security: rustls updated to 0.23.45 (RUSTSEC-2026-0285).

Installing

Download Korero_1.42.0_x64-setup.exe below, or let Kōrero update itself. The installer isn't code-signed yet, so Windows SmartScreen may warn you: choose More info → Run anyway. Your settings, meetings, notes and recordings are kept. The window opens larger once, to fit the new layout.

Ngā mihi!

Kōrero v1.41.0 — meetings that keep working, and New Zealand English on by default

Kia ora! This is the first public release since 1.28.0, so it rolls up everything from 1.29 to 1.41.

New in 1.41.0

• Meetings keep working when you leave the tab. Stopping a meeting, re-transcribing, importing, recovering a recording and generating notes now all carry on in the background. The spinner and progress are still there when you come back, and nothing you come back to can overwrite a finished result.
• Kōrero tells you when a recording isn't hearing anything. If your microphone picks up nothing for two minutes, or almost no sound is coming from your computer's audio output, you get a warning during the meeting while there's still time to fix it, not an empty transcript afterwards.
• Missed speech is rebuilt from the recording. If live transcription falls behind or hits an error, Stop now rebuilds that speaker from the saved audio.
• Clearer failures. Re-transcribe reports the real error and keeps your existing transcript. It no longer replaces it with nothing.
• Honest notes. Notes for a very long meeting say what share of the transcript they cover, and notes cut short by the model's length limit say so.
• New Zealand English is on by default. Macrons, NZ place names and NZ spelling now apply out of the box. If you had plain "English" selected, it switches to New Zealand English once. Turn it off in Settings → General if you prefer, and it will stay off.
• A word split by the speech engine ("W Nau") no longer comes out in capitals ("WHĀNAU").

Also since 1.28.0

• Meetings: stronger protection against losing meeting data; importing .m4a recordings from phones and Teams now works (ALAC audio).
• Te reo Māori: macron restoration no longer turns a singular into a plural; there's a new "Tidy te reo" action; and the macron lexicon now comes from a single source.
• Dictation: spoken formatting cues such as lists are now applied the same way every time, instead of depending on the AI model.
• Look and feel: a new app icon (a K with a macron), a quieter design, a recording overlay that's no longer cut off, one date format across the app, and a model list that shows only the models you have installed.

Installing

Download Korero_1.41.0_x64-setup.exe below, or let Kōrero update itself. The installer isn't code-signed yet, so Windows SmartScreen may warn you: choose More info → Run anyway. Your settings, meetings and recordings are kept.

Kōrero v1.28.0 — Trim meeting audio, vocabulary fix, keyboard focus

First release since v1.22.0 to actually ship an installer. If you're running an older build, auto-update should pick this one up from here on.

Highlights

Trim meeting audio. Set virtual in/out points on a meeting straight from the transcript — "Start here" / "End here" on any segment. The audio file is never modified; the markers hide segments from copy, export, search and post-processing, and one click clears them. If you generate notes and then change the trim, the notes are marked stale rather than quietly describing audio you cut out.

Learned vocabulary no longer eats ordinary English. Saying "project" could produce "ProjectIQ" — and worse, the greedy match deleted the following word, so "the project is on track" became "the ProjectIQ track". Fuzzy matching now requires a close edit distance and refuses to fire when the spoken word is ordinary English. Macron restoration and te reo terms are unaffected.

Trade-off, stated plainly: a custom term spelled like a common word (a client called "Beacon" heard as "bacon") is no longer auto-repaired by this layer. Teach a correction for those instead.

Keyboard focus ring. Interactive controls now show a visible aurora-cyan focus ring when you navigate by keyboard (~10.7:1 contrast against the navy base), plus a reduced-motion mode.

Support contact. The About page now has a Support row that opens a pre-filled email to nic@kyt.nz.

Also in this release

Rolled up from v1.26.0 and v1.27.0, which were tagged but never published:

• Sidebar now reads "Post-processing" — the earlier relabel had been inert for six weeks because i18n won over the fallback label
• System-audio dropouts explain themselves: AUDCLNT errors now name the likely cause (headphones, Bluetooth, a dock, a driver reset), say what it cost the recording, and give the recovery action, instead of a bare hex code
• Transcript segments carry timestamps

Upgrading — please read

The app's identifier changed to com.kyt.korero, which also changes where it stores your data. The app migrates this for you automatically on first launch — settings, taught corrections, meetings and already-downloaded speech models are moved across, so there is nothing for you to do.

It's a folder rename, so it's instant regardless of how many GB of models you have. If it fails for any reason (a file locked by another running copy), your old data is left untouched and simply relaunching retries.

Install

Download Korero_1.28.0_x64-setup.exe below and run it. The installer is unsigned, so SmartScreen will warn — More info → Run anyway.

Speech models download once on first run; after that transcription is entirely on-device. Your audio and text never leave your computer.

Privacy note

This build no longer carries the maintainer's build-machine paths or personal email in the binary. Contact for this fork is nic@kyt.nz.

Kōrero v1.25.0 — Safety, Accessibility & Transcription Fixes

Highlights

• Delete confirmations across History, Notes, Meetings and the recordings recovery list — no more single-click, irreversible data loss.
• Keyboard-accessible dropdowns — full arrow/Home/End navigation, Escape-to-close, and proper ARIA roles throughout the settings UI.
• Live progress feedback during meeting transcription and post-processing, with screen-reader-friendly status announcements.
• One audio at a time — starting playback anywhere in the app now pauses any other audio already playing.
• Faster empty-dictation handling — skips post-processing entirely when there's nothing to clean up.

Also in this release

• Fixed a transcription repetition bug where meeting imports could produce runaway word/phrase loops (e.g. "project, project, project…") on certain audio. A new n-gram run-collapse guard catches these at any granularity while leaving natural repeated speech ("no, no, no") untouched.
• Configurable recording & export paths — meetings can now be saved to a custom folder (Save-As dialog on every export, defaulting to Documents), with a one-off option to move existing recordings when you change folders.

Notes

No prior tagged build (v1.17.1–v1.24.0) was ever published publicly — they sat as GitHub drafts. This is effectively Kōrero's first public release. If you're running an older installed build, grab the installer below; auto-update will pick this release up from here on.

Kōrero v1.22.0

Kōrero v1.22.0

A big update across Meetings, audio briefs, and the Models page — all still fully on-device.

Meetings

• Record meetings capturing both your microphone and the system audio (calls, videos)
• Automatic date-and-time titles, and search across all your meetings
• Fix transcripts: select a word to teach a correction (fixes it everywhere and improves future transcriptions), or hover any line to edit it directly
• Turn transcripts into tidy notes, edit them in place, or give the AI feedback to refine them

Audio briefs

• Turn any notes, meeting, or text into a short spoken MP3, read aloud by an on-device voice
• Choose the voice, style, and pace
• Download the MP3 or open its folder

Models

• One place to manage every model — speech-to-text, post-processing, and the text-to-speech voice — with a live check of whether the voice engine is set up

Polish & under the hood

• Cleaner navigation and icons
• Loudness normalisation and context-biasing for more accurate transcription
• Automatic updates

Everything runs locally — your audio and text never leave your computer.

Korero v1.19.2

See KORERO_STATUS.md changelog. Headline: taught corrections now persist and apply.

Korero v1.18.0

Korero v1.18.0 -- one-click auto-update (signed, fork-locked endpoint), zero-latency dictation sequencing (a new dictation starts instantly while the previous one is still processing), dependency licence/advisory compliance green. From this version onward the app updates itself.

Korero v1.17.1

Kōrero v1.17.1

First public release. Offline speech-to-text for Windows, rebranded and re-themed fork of Handy tuned for NZ English and te reo Māori.

Highlights (v1.17.x)

New in 1.17.1

• Privacy pills in the footer: see at a glance whether audio and text stay on-device or which cloud provider AI clean-up uses
• Prompt aliases with fuzzy search: type "email", press Enter, prompt selected
• Aurora sound theme: asymmetric start/stop cues (sharp high start, soft descending stop) — opt-in under sound settings
• On-device privacy copy throughout onboarding and Help
• Security: rustls-webpki and tar advisories patched (RUSTSEC-2026-0049/0098/0099/0104, 2026-0067/0068)
• CI: GitHub Actions (clippy + test on push; release builds on tags)

From 1.17.0

• Ollama doctor: detect / install / start / self-heal for local AI clean-up
• Merge meetings (Part 1 + Part 2 imports combine chronologically)
• Imported recordings no longer warn about missing system audio

From 1.16.x

• Meeting import accepts m4a / aac / mp3 / flac / ogg (phone + Teams recordings)
• In-app update notifications (this release activates them)
• NZ English + te reo Māori blend prompt; click-any-word-to-teach corrections in History

Install

• Installer: Korero_1.17.1_x64-setup.exe (unsigned — SmartScreen will warn; More info → Run anyway)
• Portable: unzip Korero-portable-1.17.1.zip and run korero.exe

Speech models download once on first run, then transcription is 100% on-device.