Changelog
v0.1.0
New Features
- Regex-based snippet matching — Snippets now support regular expressions alongside simple literal text matching. The snippet configuration dialog has been redesigned with reorderable matchers, a type selector (literal or regex), optional regex flags, and live validation against your full snippet list.
- New "Basic" enhancement preset — A new formatting option called "Basic" focuses on fixing grammar and spelling only, without changing tone or style. It appears alongside the existing Normal, Prompt, Email, Formal, Casual, and Romantic presets.
- Profile edit and delete from the sidebar — You can now rename or delete any profile (except the Global profile) directly from the sidebar, without navigating to the profiles page.
- App-based profile auto-selection ("Default for apps") — Each profile now has a section where you can assign specific applications. When a matching app is focused, that profile is used automatically without pressing a hotkey. You can add apps directly from the profiles card view.
- System tray theme support — The tray menu icons now match your operating system's light or dark theme and update automatically when the theme changes.
- Profiles page redesign — The profiles list is now a card grid showing the formatting preset, hotkey shortcuts, and associated apps (with their actual icons). You can add apps to a profile right from the card.
Improvements
- Recording continues when the active window is unavailable — If the app can't detect which window is focused (e.g., due to missing OS permissions), recording now starts gracefully instead of showing an error.
- Smoother startup — The onboarding status check now loads asynchronously, eliminating a brief flash on startup.
- Invalid snippets are silently cleared — If your snippet configuration is corrupted or uses an incompatible format, it is automatically reset rather than causing errors.
- Tray menu cleaned up — The tray menu now has a separator between items for clearer visual separation.
Breaking Changes
- Cloud LLM-based transcription removed — The option to use a large language model directly for speech-to-text in cloud mode has been removed. LLMs are now only used for text enhancement after transcription. If you were using an LLM for cloud transcription, you'll need to select a dedicated speech recognition (ASR) model instead.
- Routing rules page removed — The separate "Routing" page is gone. That functionality is now built into each profile's settings under the new "Default for apps" section. Any previously configured routing rules will need to be recreated as app triggers on the relevant profiles.
- Profile icon picker removed — The icon customization option when editing a profile has been removed. Profiles no longer have custom icons in the sidebar.
- Snippet format changed — Snippets now use an ordered array of matchers with a
whenfield (containing literal or regex matchers) and abodyfield, replacing the previous simple shortcut-to-expansion format. Existing snippets that don't match the new format will be automatically cleared. - "Supports Audio" toggle removed from cloud model dialog — The option to mark an LLM as accepting audio input is no longer available, since LLMs are no longer used for direct transcription.
v0.0.132
No notable changes in this release.
v0.0.129
No notable changes in this release.
v0.0.127
New Features
- Translate your text into 100+ languages. A new "Translate" option is now available in the formatting choices. After picking it, you can choose a target language from a searchable list (each with its flag and native name) — from English, French, and Spanish to Swahili, Cantonese, and Hawaiian. You can search for a language either from the quick pop-up or the full picker window.
- Recent items in the system tray. The tray menu (notification area icon) now shows your 10 most recent results. Click any entry to copy its text straight to your clipboard — no need to open the app.
Improvements
- The tray menu now has an "Open OiPer" option instead of the old Show/Hide toggle, so it always brings the app window forward when you need it.
- Tray recents stay up to date. The list of recent items in the tray refreshes automatically as you generate new results, retry past ones, or delete history, so it always reflects your latest activity.
v0.0.126
Improvements
- On macOS, clicking the app's icon in the Dock now brings the main window back to the front (restoring it if it was minimized or hidden).
v0.0.125
No notable changes in this release.
v0.0.123
Bug Fixes
- The setup status panel now updates immediately when a speech model download starts, fails, or is cancelled — previously it could keep showing stale warnings until the next refresh.
Improvements
- While a speech model is downloading, the status panel now shows it as "downloading…" instead of warning that the model is missing.
v0.0.122
New Features
- Linux x64 support — The app is now available for Linux (x86_64) users. Install packages are provided in deb, rpm, and AppImage formats, with all system dependencies handled automatically.
Bug Fixes
- Overlay window now interactive on Linux — The overlay window no longer ignores cursor events on Linux, so you can click and interact with it normally. Behaviour on Windows and macOS is unchanged.
- Error messages no longer show macOS‑only instructions on other platforms — If the global hotkey listener fails, the error message will only suggest granting Input Monitoring permission in System Settings when you are on macOS. Windows and Linux users will see a cleaner message without irrelevant steps.
- Hotkey configuration on Linux uses the correct platform‑specific keys — Hotkey hold, toggle, and auto‑submit settings are now properly stored under
hotkeyHold.linux,hotkeyToggle.linux, andautoSubmit.linux, matching the pattern already used on macOS and Windows.
v0.0.121
Now I have a thorough understanding of all the changes. Let me generate the release notes.
Improvements
-
Smoother handling of empty or silent recordings — When a recording results in an empty transcription (silence, background noise, or only non-speech markers like
[Music]), the app now silently skips it instead of showing an error. This means fewer false alerts and a cleaner experience. -
Cleaner text enhancement output — The AI model used for rewriting or improving text now follows stricter output rules: it no longer adds introductory phrases such as "Here is the rewritten text" or "Sure, here's the improved version". The output is just the text itself, with no preamble.
Bug Fixes
- Fixed empty transcription errors for cloud services — Cloud speech-to-text services (both ASR and LLM-based) sometimes returned empty or meaningless results. Previously these would cause an error; now they are handled gracefully and simply ignored.
v0.0.120
Improvements
- More accurate processing time display – Processing time is now calculated by summing the actual durations of each processing step (resampling, context collection, transcription, enhancement) rather than using a simple start-to-end difference. This gives a more precise and reliable reading of how long each stage took.
- Better duration formatting for short times – Durations under one minute now display seconds with one decimal place (e.g., "1.5s") instead of showing "0m 0s", making them easier to read and compare at a glance.
- Total time in overview now includes processing – The overview cards and chart now account for processing time alongside recording time when showing total duration, providing a more complete picture of how long each session took overall.
v0.0.119
No notable changes in this release.
v0.0.116
Improvements
- Better keyboard shortcut display on macOS – On Mac, modifier keys in shortcut hints now show the correct macOS icons (like
⌘,⌥,⌃) instead of their Windows counterparts. Hotkey icons throughout the interface are also slightly larger for improved readability.
Bug Fixes
-
Clearer auto-submit description – Updated the instruction text in the settings to better explain the auto-submit behavior. The text now reads "Automatically submit after inserting the transcription" instead of the previous wording.
-
Tighter spacing for auto-submit instruction – Adjusted the alignment of the auto-submit explanation so it sits more comfortably next to its label.
v0.0.115
New Features
- Auto Submit section added to profile settings. You can now configure a key to press after transcription that automatically submits the text. This setting has been moved from the hotkey grid to its own dedicated section on the profile page, making it easier to find and manage.
Improvements
-
Hotkey labels updated for clarity. "Press & Hold" is now labeled "Push to talk", and "Toggle" is now "Press to start/stop" — making it clearer what each hotkey does.
-
Shift keys now show with an arrow icon. Instead of displaying "LShift" and "RShift" as text, the Shift keys are now shown with a compact "L" or "R" label paired with an up-arrow icon, matching the visual style of other modifier keys.
-
Better responsive layout for hotkey and language selection. The grid now adapts more gracefully across different screen sizes, ensuring the language dropdown and hotkey options remain well-organized on tablets and smaller screens.
v0.0.114
Improvements
- Refined how operating-system–specific configuration settings are handled. Settings like the whisper backend, injection method, hotkeys and auto-submit now reliably use their platform-specific keys (for example,
whisperBackend.windowson Windows orhotkeyHold.macoson macOS). If you edit your configuration file manually on Windows or macOS, make sure you are using the OS-specific key names – generic keys without the platform suffix are no longer read on those systems.
v0.0.113
New Features
- Toggle shortcuts for profiles. Each profile now supports two types of shortcuts: a "Press & Hold" shortcut (start recording while held, stop when released) and a separate "Toggle" shortcut (press once to start recording, press again to stop). You can set both or just one, depending on how you prefer to work.
- Auto-submit after dictation. After your dictated text is inserted, you can now choose to have the app automatically send it — just like pressing Enter, Ctrl+Enter, Shift+Enter, or other common submit key combinations. Configure this per profile in the profile settings.
- Press Escape to stop a toggle recording. If you started a recording with a toggle shortcut, pressing Escape will now stop it without needing to press the toggle shortcut again.
- Cancel button on hotkey fields. Each hotkey field now shows a small X button when you hover over it — click it to quickly clear the shortcut.
- Hotkey validation is now clearer. The Escape key can no longer be set as a shortcut. The app also prevents you from assigning the same key combination to both the hold and toggle shortcuts for the same profile.
- New language option: "Translate to English". The previous "Auto English" option has been renamed to make its purpose more obvious.
Improvements
- OiPer model names and descriptions refreshed. ASR models are now called "Pro" (most accurate) and "Flash" (fastest). LLM models have been renamed to a celestial theme: Auto, Earth, Moon, Mars, Jupiter, and Pluto. Each model now shows a short tagline beneath its name and a detailed description in a tooltip on hover.
- Model ratings now use a 1–5 dot scale. The quality, speed, and cost indicators for OiPer models have been simplified to show a clearer 1-to-5 rating.
- Nova model cost updated. The Nova (Flash) ASR model's cost rating has been reduced, making it a more budget-friendly option for near real-time transcription.
- Profile header simplified. Profile names now sit next to a one-line description of the mode — the mode badge icons and badges have been removed for a cleaner look.
- Doctor now warns about missing hotkeys. The Settings doctor will alert you if a profile has no hotkey assigned at all, so you never lose track of an unusable profile.
- Duplicate hotkey detection covers both shortcut types. The doctor now checks for duplicates across both hold and toggle shortcuts, and its messages clearly indicate which profile uses which type.
v0.0.112
New Features
-
Application context tracking for OiPer mode: When using an OiPer profile, the app now captures which application and window you're currently using (app name, window title, and visible text). This context is sent along with your recording for smarter processing. Context capture runs in the background so it doesn't slow down recording.
-
Application icons across the app: Application icons are now shown in several places — the Overview page's "Most Used" card displays the icon of your most frequently used app, and the History details dialog now shows the app icon next to the application name. Icon support has also been added for Windows (previously only macOS was supported).
-
Revamped Overview cards: The overview page now has a four‑column layout on larger screens. A new "Most Used" card shows which application you use most and what percentage of sessions it appears in. Typing speed now displays as "X× faster than typing" (or "Warming up") instead of a percentage comparison. All "today" stats and the "…/transcription" labels have been replaced with cleaner "avg per session" summaries.
-
OiPer profile model and template selectors: Profiles in OiPer mode now offer two ways to choose models:
- Simple mode: Pick a predefined template (Default, Fast, Accurate, Slow) — each has a description and a unique icon.
- Advanced mode: Independently choose an ASR (speech recognition) model and an LLM (enhancement) model. Each model shows quality, speed, and cost ratings as easy‑to‑read dot meters.
- The enhancement model selector is only shown when formatting is enabled, keeping the interface tidy.
-
History details improvements: The details dialog now shows:
- The application you were using (with its icon) as a dedicated field.
- How long context collection took ("Context duration").
- The Whisper backend field now appears earlier in the metadata list.
- For OiPer mode recordings, transcription and enhancement are shown as a single "Processing" step in the timeline instead of two separate steps.
- A "Preparation" step has been added to the timeline, and all timeline colors have been refreshed for better readability.
- Duration formatting is now more human‑readable (e.g. "1 minute 30 seconds").
v0.0.111
New Features
-
New text enhancement presets: Casual and Romantic. You can now choose "Casual" to rewrite your text in a friendly, conversational tone, or "Romantic" to give it a warm, affectionate feel. All existing presets (Normal, Prompt, Formal, Email) are still available.
-
Task-based custom formatting. Instead of writing a full system prompt, you can now describe a simple task — for example, "translate to French" — and the AI will automatically apply a formatting template to produce the result. A new "Task" option appears alongside "Custom" in the formatting selector.
-
Toggle between Task and Advanced mode. When using custom formatting, a checkbox lets you switch between Task mode (describe what you want) and Advanced mode (write the full system prompt directly). A tooltip explains the difference at a glance.
Improvements
-
Better save feedback. The Save button now shows a floppy disk icon, and changes to a green checkmark with "Saved!" after you save — so you always know your changes have been stored.
-
Cleaner formatting selector. The preset buttons now use consistent icons throughout, making it easier to identify each option at a glance.
-
Sharper AI prompts. The text sent to the enhancement LLM is now structured more clearly, which helps the AI follow your formatting instructions more accurately.
Bug Fixes
- Fixed an issue where the enhancement system would fall back to the transcription model when no enhancement model was set. It now correctly uses only the enhancement model you selected.
v0.0.110
New Features
- Delete transcription records from the details dialog – You can now remove a transcription from your history directly from the details view. A confirmation prompt asks before deleting, so accidental removals are avoided.
- Navigate to a profile from the details dialog – Profile names in the metadata section are now clickable links that take you directly to the corresponding profile page.
Improvements
- Timeline redesigned as a visual progress bar – The processing steps (resampling, transcription, enhancement, injection) are now shown as a horizontal, color-coded progress bar with a legend underneath. This makes it easier to see at a glance how long each step took relative to the total.
- Smarter output display – The details dialog now shows the final output (the best available result). Separate transcription and enhancement sections appear only when they differ from the final output, reducing clutter when they are the same.
- New metadata fields – You’ll now see Recording duration and Processing time directly in the details. The Whisper backend used for transcription is also shown.
- Cleaner metadata layout – Information is presented in a compact two-column grid with clear labels, making it easier to scan.
- Renamed “Input Audio” to “Recording” – The audio player section now has a clearer label.
- Close button in the dialog footer – A dedicated Close button has been added to the bottom of the details dialog for convenience.
Removed
- The following fields are no longer shown in the details dialog: Status, Started, Completed, Duration (overall), Enhancement Method, and Enhancement Tech. This information was either duplicative or not commonly needed.
v0.0.109
No notable changes in this release.
v0.0.108
After carefully reviewing all commits, these changes are all internal build system and developer tooling adjustments:
- Commenting out CMake environment variables in the Makefile and release workflow (affects compiler flags during build, not runtime behavior)
- Simplifying a build log message in the Makefile
- Removing a custom build variable that was only set during local development (never used in release builds)
- Adding a
.envrcfile for developer environment automation - Reordering and adding CPU feature flags for local Windows development builds (release builds already had these)
None of these changes introduce new features, fix user-facing bugs, or alter the application's behavior for end users.
No notable changes in this release.