SpeechKit v0.70.0

Highlights

• A clearer Library. Transcriptions, Quick Notes and Voice Sessions have
  their own tabs. Browse further through your history, expand and copy entries,
  and pin a transcription you want to keep close.
• More dependable first runs on macOS. SpeechKit selects the audio backend
  for your platform instead of starting with a Windows-only setting.
• Easier release tracking. Changes now arrive in named release packages
  with compact patch numbers and a shared summary across distribution channels.

Added

• library: the Library now gives each mode its own tab — Transcriptions,
  Quick Notes and Voice Sessions — instead of three columns sharing one screen,
  and each tab loads further back through your history a page at a time rather
  than stopping at the newest twenty entries. Any row expands in place so you
  can read and select the whole text, every row copies, and a transcription can
  now be pinned the way a quick note already could, so a line worth keeping
  stays at the top however far back the history has moved. A search box filters
  what is currently loaded.
• integrations: integration cards can show real brand logos, served from a
  logo CDN, when the build is configured with a public client id. Without one —
  and under the "Local network only" and "This device only" privacy scopes,
  where no request may leave the machine — every card keeps its bundled logo and
  nothing is fetched.

Changed

• settings: Mode Settings now sits above Audio Settings in the sidebar, so
  you choose what SpeechKit does before adjusting how it sounds.
• settings: the kombify Cloud card no longer overflows. Its explanation of
  why a build cannot sign in, and the list of modes it serves, moved into the
  account dialog; the card keeps a single-line status and shows Connect only
  when it can actually reach the sign-in page. If your browser did not open, the
  dialog can now reopen the sign-in page instead of leaving you to copy the URL.

Fixed

• macOS: a fresh install could not start at all. The first run wrote a
  configuration naming a Windows audio backend, the app refused to open a
  microphone with it, and — because logging is off by default — it exited
  without saying anything: no message, no log, no crash report. The audio
  backend now resolves per platform, and continuous integration launches the
  built macOS app on every change so a first run that cannot start is caught
  before a release carries it. Windows is unaffected; it resolves to the same
  backend it always used.

• macOS: the app's global-hotkey listener is installed but receives no key
  events, so the Dictation, Assist and Voice Agent hotkeys do not respond. This
  is not fixed yet. It is listed here because the macOS build is a beta and the
  behaviour is now known rather than suspected: hold-to-talk on macOS does not
  work in this build, and the menu-bar item is the way in until it does.

• overlay: the hover strip and the radial menu no longer replace their
  controls with "Input submitted; app receipt unconfirmed" after a dictation
  that simply worked, and no longer keep doing so on every hover until the
  next dictation. The notice now appears only when text needs rescuing:
  it was not inserted, may have been inserted partially, or could not be
  saved to the history. It stays while you retry and closes when dismissed.

• dictation: pasting into a window that had moved its focus to one of
  its own dialogs or popups no longer fails with "not inserted"; the window
  counts as in front as long as the dialog belongs to it. A maximized target
  is no longer shrunk to its windowed size while dictation brings it back to
  the front, a window that was closed while you spoke is reported as gone
  instead of being fought over, and a recording started right after a click
  on the overlay pill says so rather than pasting into the pill. The "not
  inserted" notice now names the reason: the window closed, it was not in
  front, or the overlay had focus.

Full Changelog: v0.69.41...v0.70.0