Release notes

Changelog

What’s new in Hyppo · current 0.1.3.

[Unreleased]

Small Whisper and NVIDIA Nemotron 3 Diarization download on first launch
Live transcription while audio is recording (screen, AI capture, and meetings); audio-only takes show the transcript in the app window as speech is recognized. When the take stops, Whisper and Nemotron re-run on the saved file so the stored transcript is not the live 5-second windows
Video-call transcripts and summaries use participant names from the calendar invite and the call window
Google Calendar meeting guests are loaded for Meet transcripts (invite list, including a full event fetch on save)
Microphone level is tracked so your own speech can be labeled in the transcript
Google Meet windows are scanned for the people list and the moving blue speaking ring so transcripts can name who is in the call and who is talking
Settings Appearance Light / Dark mode for the desktop app chrome
Add to Slack button on Settings and share, using the Hyppo Slack app
Shared UI primitives: Button, IconButton, Icon, Field, ChoiceRow, Modal, and ConfirmDialog
Dedicated macOS template tray icon (and color tray fallback) for the menu bar

Marketing site palette matches the desktop app dark theme (indigo, blue, purple)
Marketing site header, footer, and favicon use the vector hippo mark
Free tier cap is 2 minutes per video
Plan names are Free, Personal, Pro, and Platinum
Upgrade is hidden in the app on Pro and Platinum
Audio-only recordings skip the live screen mirror in the app window
App chrome, settings, library, editor, and dialogs use the shared UI components
Brand mark is smaller in the top bar; meeting bar uses a dedicated mark style
App icon refresh; production site defaults and contact email use hyppo.run

Saved transcripts no longer keep live Whisper windows (which skipped Nemotron and hallucinated `[BLANK_AUDIO]`)
Google Meet OCR ignores shared-tab chrome, so ads, cookies, and Stop presenting are not saved as people or speakers
Transcripts put each new speaker on a new line
Regenerating a meeting summary no longer overwrites the transcript
Library thumbnails and durations persist across restart (sidecar next to each video, stem-named thumbs)
Library length for Chrome WebM takes (no duration header) is read from ffmpeg packet timestamps
Whisper Medium q5 install downloads `ggml-medium.en-q5_0.bin` (the q5_1 file 404s on Hugging Face)
Meeting summary regenerate finds disk-only meetings; looped Whisper transcripts are collapsed
Quit Hyppo ends Electron and, in `npm run dev`, Vite as well so port 4344 is freed
Meeting bar Stop, Cancel, and Minimise labels stay centered when the meeting name is long

Supabase accounts on the desktop app and marketing site (email, Google, browser handoff), device activation, and Stripe checkout / billing portal edge functions
Meeting detection: prompt to record from calendar lead/active windows and live Zoom / Teams / Meet / Webex titles
Marketing site pages: pricing, FAQ, why, account, login/signup, auth callback, changelog, contact, privacy, terms, download
Mic and webcam device pickers in Setup; export and share actions in the editor top bar
`Makefile` targets for common run / test / build / site / release workflows
Optional Google OAuth client secret for Web clients (Desktop PKCE still works without it)

Empty takes when the headless toolbar was fully hidden (now stays shown at opacity 0 so capture streams keep running)
Finish/commit refuses a missing or zero-byte recording file instead of publishing an empty object
Overlay draws: compact control bar stays out of capture; full-screen draw mode turns protection off so strokes appear in the take

App chrome and recording UI restyle (frameless macOS title bar, logo mark, settings cards)
Site infrastructure defaults to `eu-west-2`; GitHub OIDC defaults to `sourcetown/hyppo`
Renamed product from WindUp/WUP to Hippo (`window.hippo`, `hippo://`, paths, installers)
Removed Upvolt Central sign-in; Slack share uses a local Slack app only

Recording overlay Pause / Draw / Stop respond again after countdown (click-through hit-test used geometry instead of event.target)

Build Windows installers with `scripts/Build-Win.cmd` (mirrors WSL UNC checkouts to a Windows build dir)
Build Windows and macOS installers on local machines; tag CI runs tests only (no electron-builder on GitHub runners)

Meetings view: join upcoming calendar events, capture system + mic audio, and save past meetings
Google Calendar and Outlook OAuth connect in Settings, with merged upcoming event lists and join links
On-device Whisper transcript, seeded notes, and optional OpenAI-compatible meeting summaries after capture
Per-meeting keep flags for audio, transcript, notes, and summary (defaults configurable in Settings)
Keep a Changelog workflow: `CHANGELOG.md`, `npm run release` / `release:check` / `release:notes`, and CI notes on GitHub releases
Download site “What’s new” section fed from changelog notes in the release manifest

Local screen or window recorder with a library shell
Local SQLite catalogue (optional Postgres) with titles, tags, search, and transcripts
On-device Whisper transcription installed from Settings after first launch
Slack share via a local Slack app, including self-DM fallbacks
Record for AI: stills, audio, and a tagged transcript under Videos/Hippo/AI
Painted click circles, mouse zoom, and timestamped speech in finished takes
Multi-platform installers (Windows, macOS, Linux) with S3 + CloudFront download hosting
Download website driven by the release manifest and CloudFormation stack outputs