Version journal

Dictation and hover translation now work in Electron apps (Slack and similar) even when they don't.

• Fixed dictation silently failing to insert text in some Electron apps (Slack confirmed) on some Macs: when accessibility can't identify exactly which app is focused, the app now sends the keystroke the same way a real keypress would travel — to whatever is actually focused — instead of giving up and leaving the text only in the clipboard.
• Hover translation now wakes the target app's accessibility tree before reading the text at the cursor, the same way dictation already did. It didn't before, which is the likely reason it sometimes found a container with no readable text in Electron apps.
• The three dictation modes and the secretary now lock each other out: starting one while another is recording does nothing, instead of running two sessions on the same microphone at once.
• Added a "launch at login" switch (Settings → Subtitles & translation).
• The dictation panel now says what it's doing — "Слушаю и перевожу…" for translated dictation, "Слушаю и записываю за вами…" for the secretary — and widens itself to fit instead of cutting the text off.

A "Permissions" screen that tells you exactly what's granted, instead of guessing.

• Added a Permissions window (menu bar icon → Permissions…) that shows the live status of everything the app needs — microphone, speech recognition, Accessibility, and app-audio capture — each with a one-click button to request it or jump straight to the right System Settings pane.
• App-audio capture has no way to be checked in advance on macOS; the window now has a Test button that makes a real, harmless attempt right there, instead of requiring a full listening session just to find out whether it's allowed.
• The menu bar now warns right at the top when something needed is missing, instead of failing silently later.
• Fixed a misleading message during dictation: it used to guess "no Accessibility access?" even when Accessibility was already granted, sending people through pointless permission resets. It now checks the real state and logs the actual reason.

Dictation now inserts reliably everywhere, and every hotkey in the app can be rebound.

• Fixed dictation not inserting text in Electron apps (Slack, Discord, VS Code and others) while a non-Latin keyboard layout was active — the app now switches to a Latin layout for the instant it sends the paste keystroke, then switches back.
• Every hotkey-bound action can now be rebound: click into the field in Settings and press the combination you want, instead of picking from a fixed list. Covers all three dictation modes, copying the last dictation, holding to see a hover translation, toggling hover translation on and off, and starting or stopping subtitles. A sequence already taken by another action is flagged right there.
• The three dictation modes now show up as clickable items in the menu bar, next to the key combination that triggers them — click instead of remembering the shortcut.
• Settings are split into three tabs — Dictation, Secretary, and Subtitles & translation — instead of one long scrolling form.
• The secretary's voice-command list is now an editable table: add, remove or rename commands, instead of only switching the built-in ones on and off.
• Faster startup and lower memory use: the speech-recognition and pause-detection models load once and are shared between dictation modes, instead of loading twice.

Dictation, in three modes.

• Press ⌥Space, speak Russian, get back exactly what you said, typed where the cursor is.
• Press ⌥⌃Space instead and get the English translation, typed the same way.
• Press ⌥⌘Space for a secretary mode built for long dictation sessions: text is inserted piece by piece as it's recognised, and it keeps writing into the field where the session started even if you switch to another app while still talking.
• Dictation runs through a separate on-device speech model, chosen because it keeps punctuation — Apple's own Russian dictation engine does not.
• A small panel shows the session as it happens. Drag it by its background and it stays where you left it.
• A word-replacement dictionary fixes names, company names and terms that speech recognition keeps getting wrong. One button copies the last dictation to the clipboard.

The first release.

• Subtitles with a Russian translation above all windows. Listens to the sound of a chosen app or to a microphone; English speech is recognised and translated on the machine itself.
• Show the English line, the Russian one, or both at once — switched with buttons on the panel.
• Two ways to show the text: only the current phrase, or everything one phrase after another, with scrolling back.
• Font, size and colour set separately for the English and the Russian line, and background opacity on a slider.
• The panel is dragged and stretched with the mouse; its position and size are remembered.
• The last 50 sessions stay in «Недавние транскрибации» (Recent transcriptions), and the text is copied with one button.
• Translation of the text under the cursor: hold ⌥⌘ and point the mouse at English text in any app.
• Together with the translation, the macOS system dictionary entry for the word under the cursor.
• The pin turns the panel into a separate window: the original can be edited, the translation is recomputed, and every block can be read aloud.
• Selected text is taken whole, even when the selection covers a heading and several paragraphs, and bold and italics are kept.