Skip to content

Hexium 3.3.0 (783bb3e)

Choose a tag to compare

Hexium 3.3.0, built on Hex 3.3.0 (67bc1cf) from 783bb3e.

Install: Unsigned, so Windows SmartScreen and macOS Gatekeeper ask once. Windows: run hexium-setup-3.3.0-windows-x64.exe (or -arm64); macOS: open the DMG and drag Hexium to Applications. This build does not update itself; later releases do.

Hexium

  • Updated: Build against patched transcribe.cpp and Rune
  • Updated: Own identity so it installs beside Hex; no self-updates
  • Updated: Indic-Transcribe-Flex as a built-in model
  • Updated: Update notice
  • Updated: Auto language for Indic-Transcribe-Flex
  • Updated: Learn from your fixes and quote what's on screen

From Hex (76)

  • Warm speech models on silence without the vocabulary prompt (#1569)
  • Stop launching osascript on every context poll (#1567)
  • Linux: let the copy land before reading the clipboard in typed text (#1563)
  • Fix windows meeting reveal selecting the exported file (#1558)
  • Ignore microphone dc offset in level meters (#1560)
  • Bump rune: run a due timer covered when its frame never comes (fixes 100% cpu with a covered settings window on wayland) (#1555)
  • Labs: release notes link in the sidebar footer (#1554)
  • Offer auto for models that detect language themselves (#1551)
  • Onboarding: only offer a language switch when the system has one (#1550)
  • Windows: keep dictation on clipboard and copy dictation only labs (#1546) (#1549)
  • Hud: release the painter, and with it the GPU device, once the HUD hides (#1542)
  • Unload idle: start with the model unloaded (#1544)
  • Add orukeet to the speech model catalog (#1545)
  • Cli: select the model after an offline model install (#1443)
  • Labs: terminal paste shortcut on Windows and Linux (#1534)
  • Labs: caps lock light waits for audio, optional slow blink (#1535)
  • Feedback: release the macOS tone output soon after the last tone (#1531) (#1533)
  • Labs: Transform typed text shortcut (Linux) (#1525)
  • Linux: Annotate on GNOME lab (capsule and ink float on XWayland) (#1529)
  • Make pane scroll bars visible and easier to grab (#1523) (#1530)
  • Fix anomalyco#1518: detect a digitally silent meeting microphone (#1528)
  • Reach the Labs sidebar row with the arrow keys (#1527)
  • Fall back to the whole clip when a stream loses its opening or switches language (#1520)
  • Linux: open the annotate pen on the recorded output (#1515)
  • Keep capture media safe when Linux retention can't trash it (#1516)
  • Capture: Shift in the Annotate list marks the row under the pointer (#1503)
  • Onboarding: hold the practice invitation until the listener is up (#1511)
  • Linux: adopt the wayland display behind an xwayland DISPLAY (#1508)
  • Windows: /dictations records with Hex's own microphone, and a dictation can stream its live text (#1474)
  • Linux: adopt the Wayland display for hex dictate and fix its ready line (#1506) (#1507)
  • Labs: caps lock dictation (macOS) (#1502)
  • Linux: release pressed keys when a paste chord fails partway (#1500)
  • Linux: pad short mic gaps and reopen digitally silent streams (#1499)
  • Show why a dictation failed (#1494)
  • Linux: stopping hex always finishes, even if a quit is lost (#1495)
  • Only remove filler words from english dictation (#1492)
  • Meetings: labs toggle to name exports after the meeting title (#1493)
  • Linux: name the missing wl-copy before a Wayland paste fails (#1491)
  • Windows: hidden start makes no Direct3D device, so NVIDIA is less likely to take Hex for a game at launch (#1308) (#1487)
  • Treat remapped caps lock as a side-less modifier (#1490)
  • Settings: never replace settings.json that isn't valid JSON (#1488)
  • Linux: live typing through the wayland input method (labs) (#1485)
  • Always offer reveal files for meetings (#1483)
  • Notch indicator works with the bottom HUD position (#1466) (#1486)
  • Meetings: expire audio separately (labs) (#1479)
  • Lower audio faster and drop duck pre-roll (#1482)
  • Linux: dark tray mark on light panels (#1481)
  • History: show dictation stats in a dialog from a Stats button (#1478)
  • Labs: ⌘Q closes Settings (#1480)
  • Meetings: import audio files (labs) (#1462)
  • Fit the capture capsule's notch shell to the measured notch (#1461)
  • Labs: settings search (#1446) (#1459)
  • Labs: more call apps for meeting detection (#1458)
  • Reveal exported meeting markdown (#1457)
  • Windows: keep settings saved during dictation until the listener is idle (#1456)
  • Switch-panes waits for the window to go idle between pane clicks, untimed (#1447)
  • Keep the done tone after a delayed stop tone (#1445)
  • Idle window-open waits for General's switches before its runs (#1438)
  • Meetings: one Summaries picker for titles and summaries (#1425)
  • Log a microphone that stays gone once, not on every retry (#1427)
  • Capture routing: a later answer never moves a listed session row (#1433)
  • Keep Windows logs and models in Local AppData (#1432)
  • Meetings: platform icons always on; graduate the Meeting platform icons lab (#1426)
  • Read the window title only for script transformations (#844)
  • Tighten only the real logs folder, and never fail to start over it
  • Move the fs import into the tests
  • Make the logs folder and its files owner-only
  • Keep dictated text out of the event log (#1423)
  • History: one Kept menu with retention, Keep audio too and Clear all history… (#1416)
  • Onboarding: say the mic stays ready, or opens only while dictating (#1420)
  • Picker: one action row per model, no labels that repeat the bold value (#1417)
  • Cut the Collapsible meeting sections lab (#1406)
  • Menu bar: meeting first, then actions, then the app; Speech Model; shortcuts shown (#1409)
  • Fix anomalyco#1392: background polling and idle CPU usage (#1410)
  • Labs cards show their own settings; Voice Filter and Playground are dev-only (#1400)
  • Fix anomalyco#1402: foreground context capture and accessibility (#1407)

Every upstream commit: anomalyco/hex@83dcd39...67bc1cf