Changelog

3.1.2

2026-06-17

Fixed

• The translation language picker now works with an external keyboard (#480). The pre-take language picker could stay shut when an external Apple keyboard was connected, because the recording hotkey was tied to one physical side of the key (for example Left Control) while an external keyboard reports the other side. The hotkey picker in Settings > Recording now offers "Control (either side)" and "Option (either side)", so the same shortcut works on your built-in and external keyboards alike. Your current choice is left as it is; switch to an either-side option if you move between keyboards.
• Smoother language scrolling with a Magic Mouse (#480). Flicking the wheel on a Magic Mouse could race through the language list, because its scroll steps are far larger than a trackpad's. Each step is now capped so the list moves one language at a time, and a new "Scroll sensitivity" slider in Settings > Translation lets you tune the pace. The built-in trackpad feels the same as before.

3.1.1

2026-06-16

Added

• Translate into English (#477). English is now one of the languages you can add to your Output languages list in Settings > Translation, so you can speak another language and have VoxChimp translate it into English. It is a normal translation target, separate from the Original entry that means no translation. You no longer need to set a source language by hand: VoxChimp resolves it for you, including when your recognition language is set to auto-detect.

Fixed

• Steadier language picker (#478). When you scroll to pick a translation language in the recording bar, the wheel now advances one language per deliberate scroll instead of skating through several at once, and it stops at the first and last entries rather than wrapping around. It is much easier to land on the language you want.
• The recording bar no longer crops with a long language tag (#479). With translation on, a long sentence in the recording bar now truncates with an ellipsis instead of cutting the bar off at both edges. The full text still appears when you hover.

3.1.0

2026-06-11

Added

• Pick your translation language before you speak (#466). While idle, hold Control and press your recording hotkey to open a language picker in the recording bar, then scroll with the mouse wheel or two fingers on the trackpad to choose the output language. Your next take records in that language and the choice sticks until you change it. Settings > Translation now has a single Output languages list (up to eight) that replaces the old Target language dropdown, starting with an Original entry that means no translation at all. The separate Translate to English toggle is gone; pick English in the list instead. The picker appears once translation is on and your list has at least two languages.
• Agent Mode stays on your Mac by default. A new "Keep Agent Mode on-device" switch in Settings > Agent locks the agent to local providers (Ollama or LM Studio), so the text it reads never leaves your Mac. It is on by default; switch it off to use a cloud provider with your own API key. The note at the top of the pane now also tells you plainly, for the provider you have picked, whether your text leaves the device.
• Start dictation or the agent from the menu bar. The VoxChimp menu now begins with Start Listening and Ask Agent, so you can record or run the agent without touching a shortcut key. Each becomes Stop Listening or Stop Agent while running, and Ask Agent only appears once Agent Mode is set up.

Changed

• Settings show only what matters. The Agent pane keeps its configuration hidden until you switch Agent Mode on, and the voice output details stay tucked away until "Speak agent responses aloud" is on. The Web Search pane opens with a short note explaining that web search is optional and mainly helps a local agent reach live results.

3.0.2

2026-06-10

Fixed

• Live typing at the cursor is more faithful to what you said (#470). When streaming words live at the cursor, VoxChimp no longer types phrases hallucinated during pauses, and the last few words of a take are no longer occasionally dropped. The live typer now repairs chunk boundaries and recovers the held-back tail, so the text at the cursor matches the final transcript.

3.0.1

2026-06-08

Added

• Name the people in a labelled recording (#465). Building on the speaker labels from 3.0.0, you can now turn "Speaker 1" into a real name. Open a labelled recording in History, tap a speaker, and type their name. It then shows everywhere that speaker appears, including when you copy or paste the transcript. If you run a local AI agent (Ollama or LM Studio), VoxChimp can also suggest the names for you by reading who was introduced in the recording. That runs entirely on your Mac and the transcript is never sent to the cloud, so it needs a capable local model. Turn it on in Settings > Recording, under "Suggest speaker names automatically". Without a local agent, you simply rename by hand.

3.0.0

2026-06-08

Added

• Speaker labels (#459). VoxChimp can now work out who said what and label each speaker in a recording, all on your Mac. Turn it on in Settings > Recording > Label speakers (it needs "Save audio recordings" switched on and a one-time speaker-model download). You can copy a transcript with speaker labels from History, choose the label layout (a new line per speaker, or one inline block), and optionally have the labelled text pasted for you. It works best on clear, in-person, turn-taking audio.
• Free on-device voices for read-aloud (#459). A new neural text-to-speech option runs entirely on your Mac with no API key, and the text never leaves the device. Choose it in Settings > Voice output and download a voice model once. It joins the existing Apple and cloud voice options.
• Re-process a saved recording (#459). If a recording came out incomplete, or a stand-in engine handled it while a model was still downloading, you can now re-run it through transcription from its saved audio to recover the text, straight from the History detail view. Revert puts the original back.
• Choose how transcription runs (#459). A new Compute option in Settings > Transcription (Auto, GPU, or CPU). If long recordings on larger models lose chunks of audio, switching to GPU avoids a Neural Engine timeout that could silently drop speech. VoxChimp now also flags a recording that looks incomplete and suggests trying GPU.

Changed

• Rebuilt on the Argmax open-source Swift SDK (#459). The on-device transcription engine has moved to Argmax's unified open-source SDK, which is what makes the new speaker labels and on-device voices possible. It also brings quality fixes: fewer repeated-word loops, fewer stray phrases hallucinated during silence, and speaker-view text that now follows your spelling settings like the main transcript does.
• Custom vocabulary no longer risks dropping a spoken word (#459). Your vocabulary still corrects spellings as before, but the acoustic biasing that could occasionally swallow a word you actually said is now off by default. You can switch it back on in Settings > Post-processing if you want the extra help with rare terms.

2.20.2

2026-06-02

Fixed

• Connecting an OpenAI key now works when the key is valid (#456). The connection test in Settings > Skills could show a red "invalid configuration (HTTP 400)" for a perfectly good key. It now connects whenever your account has a usable model. When it genuinely cannot connect, because the chosen model is not on your account or the account has no credit, you get a clear amber message that explains the real reason instead of one that implies the key is broken. The test also respects the model you typed into the Model field, and the underlying provider error is now recorded in the log for support.
• Stray transcription labels no longer leak into your text. A long pause could leave a literal "[ Silence ]" in the transcript, and background-sound labels such as "[BIRDS CHIRPING]" or "[MUSIC PLAYING]" could slip through. These are now removed, while clinical abbreviations such as "[HIV]" and "[CPR]" are left untouched.

Changed

• New default OpenAI model. Fresh setups now use gpt-5.4-mini by default, with gpt-5.1 as an automatic fallback if the default is unavailable. If you have set your own model in Settings, your choice is unaffected.

2.20.1

2026-06-01

Fixed

• Silent recordings no longer paste stray text. Pressing record and saying nothing could paste your custom vocabulary list, or a stray phrase such as "Thank you.", at the cursor. A recording with no speech now produces nothing at all: no paste and no history entry. This applies to every transcription engine.
• The end of your dictation no longer goes missing. In stream-to-cursor mode the last few words could show in VoxChimp's own history but never reach the app you were typing into. The full transcript now lands at the cursor.

2.20.0

2026-06-01

Added

• Hands-free toggle mode (#450) -- you no longer have to hold the hotkey down for the whole recording. Press once to start, press again to stop. Handy for longer dictation, and you can assign more than one key to it in Settings.
• Many more recognition languages (#397) -- the recognition language picker now covers the full set Whisper understands (around 100 languages) in a searchable list, so you can pick exactly what you speak instead of relying on auto-detect.
• Translate to English as you dictate (#397) -- a new option turns your speech straight into English while it transcribes, entirely on your Mac. It works when WhisperKit is your transcription engine and sits in the Translation settings, next to the existing translation toggle.

Changed

• Smoother live typing in stream-to-cursor mode (#453) -- when live transcription types into the active app as you speak, the words now only ever move forward instead of rewriting themselves, so you no longer see text flicker or jump backwards.
• A clearer recording bar (#411, #418) -- the bar now sizes itself to its content, shows the named stage it is in instead of a vague "Processing...", and displays the full text of any error, with a button straight to the right System Settings pane when a permission is missing.
• Onboarding polish (#413) -- clearer wording, more readable text in the setup modals, a visible step counter through the three permission screens, and the setup steps now name your actual hotkey.

Fixed

• Live cursor no longer freezes on long takes (#421) -- a long recording in stream-to-cursor mode could lock up the live typing. It now stays responsive however long you speak.
• No more dropped audio while a model loads (#426) -- if a transcription model was still loading when you started speaking, part of the recording could be lost and the cursor could freeze. Model loading is now handled one at a time and held back until your recording finishes.
• No duplicated text when translating in cursor mode (#429) -- live cursor dictation could repeat or duplicate words when translation was switched on. Fixed.
• Translation failures now explain themselves (#436) -- when Apple Translation cannot run because the language pack is not installed, VoxChimp now tells you what happened and offers to download the pack, instead of quietly doing nothing.

2.19.0

2026-05-21

Added

• See your words as you speak (#396) -- VoxChimp can now show a live transcript while you hold the hotkey, instead of waiting until you let go. Choose how it appears in Settings > Recording > Live transcription: stream straight to your cursor, preview your latest words inline in the recording bar, show the full transcript in a floating panel above the bar, or turn it off. Live transcription uses Apple's on-device Speech Recognition, so your audio never leaves your Mac; languages without an on-device model quietly fall back to the waveform display for that recording.
• Stream to cursor mode (#396) -- the new default for fresh installs types your words directly into the active app as you speak. When you release the hotkey, the typed text is reconciled with the final transcript, so grammar fixes, spelling normalisation, and translation still apply. If you already use VoxChimp, your current setting is left untouched.

Fixed

• Silent ElevenLabs voice failures (#396) -- when ElevenLabs text-to-speech could not play (an invalid API key, no remaining credits, a rate limit, or a voice your plan does not include), the app used to do nothing and leave you guessing. It now shows a clear message explaining what went wrong and how to fix it.

2.18.0

2026-05-20

Added

• SearXNG setup help in Web Search settings (#322) -- when you pick SearXNG as your search provider, an info button next to the Endpoint field opens a popover with the `docker run` command to start it locally, plus a Copy button.

Changed

• Friendlier defaults for new installs (#322) -- new users now start with Insights and Voice Health visible in the sidebar, the menu-bar icon showing, and the app set to launch at login. Anyone who had already switched these off keeps their choice, and all of them can still be changed in Settings.
• Sensible first-run defaults when using Apple Speech (#322) -- if Apple Speech is the active engine on first launch, grammar fixing and audio saving start on, and spelling follows your region (for example British English in the UK). Choices you have already made are never overwritten.
• Refreshed AI provider models (#394) -- the default model for each provider in Settings > Skills now points at a current model. Anthropic, OpenAI, Microsoft Copilot and Grok previously defaulted to models that have since been retired, which could make a skill fail with a "model not found" error. Custom models you typed in yourself are left untouched.

Fixed

• Empty or cut-off transcriptions on real-world audio (#394) -- WhisperKit now splits recordings at natural pauses and is less quick to discard speech it judges "silent", so podcast-volume and multi-speaker recordings no longer drop to a few words or nothing at all.
• Whisper models failing to load (#394) -- model files no longer leak into `~/Documents/huggingface`, which on some Macs triggered a permission error that stopped Whisper models from loading after an update or rebuild.
• Wrong storage figure after switching models (#394) -- the "Storage used by models" panel in Settings > Transcription now recalculates straight away when you switch models, instead of showing a stale number for the rest of the session.

2.17.0

2026-05-18

Added

• Drag-to-Applications guidance in the DMG installer (#368) -- the DMG window now shows a cream background with a curly orange arrow pointing from VoxChimp to the Applications shortcut, plus the headline "Drag VoxChimp into Applications to install." First-time installers can see at a glance what to do, instead of guessing.
• Animated post-download install guide (#366) -- after clicking a download button on the website, an animated modal walks you through the three-step install (Open DMG, Drag to Applications, Eject). Pure-CSS animation with a `prefers-reduced-motion` fallback that shows static illustrations.
• Trial extension support (#382) -- VoxChimp now honours an extended trial period if support has granted you extra time. Your trial expiry refreshes automatically the next time the app checks in with our servers.

Fixed

• "Manage subscription" buttons silently failing (#389) -- the buttons on the active-licence screen now show a clear alert and write a log entry when the Stripe customer portal cannot be opened, instead of failing silently. Every failure path (API error, malformed URL, browser refusal) is now visible to you and auditable in `~/.voxchimp.log`.

2.16.0

2026-05-14

Added

• Voice output (text-to-speech) for agent responses and translation -- VoxChimp can now read agent (skill) responses and translated text aloud after they land. Two independent opt-in toggles control it: "Speak agent responses aloud" in Settings > Skills > Voice output, and "Speak translated text aloud" inside the Translation settings section. Both default to off.
• Four voice providers -- Apple (on-device, free, the default), OpenAI TTS, ElevenLabs, and Google Cloud Text-to-Speech. Cloud providers are BYOK; paste an API key, click the "Test connection" button to confirm it works, and switch the provider picker. The app falls back to Apple automatically when a cloud provider has no key, so voice never silently fails.
• Voice playback panel -- a small floating panel appears at the bottom of the screen whenever speech is playing, with Pause / Resume / Stop controls. It disappears when audio ends, you stop it, or you start a new recording.
• Per-provider voice and model pickers -- pick from Apple's installed voices, six curated OpenAI voices (alloy / echo / fable / onyx / nova / shimmer), six ElevenLabs shared-library voices (Rachel / Bella / Domi / Adam / Antoni / Josh), or six Google Neural2 voices across English locales. OpenAI exposes a model toggle for Standard vs HD; ElevenLabs picks between Multilingual v2 (best quality) and Turbo v2.5 (lower latency).
• Custom voice ID for ElevenLabs -- a text field under the Voice picker lets you paste a cloned-voice ID from your own ElevenLabs account; when present it overrides the curated picker selection.
• Rate control and Test voice button -- a single slider controls speech rate; the Test voice button speaks a sample line so you can verify provider + voice + rate without running a real skill.

Changed

• Simpler first-run setup (#371) -- the "Choose a transcription model" screen has been removed from onboarding. New users now see a fixed 8-step wizard, and the default Small Whisper model downloads quietly in the background on first launch (Apple Speech handles dictation in the meantime, so you can speak straight away). Model selection moves to Settings > Transcription, where it's always been available.

Removed

• Unused skill-mode overlay code -- the legacy `SkillModeOverlayController`, `SkillModeOverlayView`, and `SkillModeOverlayControllerProtocol` had no production callers and have been deleted along with their two tests. The recording bar (`RecordingBarView`) and the new voice-playback panel cover the user-facing surfaces.

2.15.0

2026-05-09

Changed

• Permissions live on the Dictate tab (#364) -- the permissions status bar now sits where you actually start dictating, instead of being buried in Settings. When microphone, accessibility, or input monitoring is missing or revoked, you see the warning where it matters and can re-grant in one click.
• Simpler Settings (#364) -- removed the standalone Permissions pane. The Dictate tab is the single source of truth for permission state, so there is no more "everything is fine" copy in Settings while Dictate quietly says otherwise.

Fixed

• Sign-up and licence session handshake -- a server-side authentication mismatch could reject some sign-up and licence-session requests. The handshake is now consistent on both ends so trials, activations, and checkout flows go through reliably.

2.14.2

2026-05-07

Fixed

• Trial, activation, and checkout -- Pro builds could be rejected at the server with an authentication error, blocking new trials and purchases. Both sides of the handshake have been corrected so requests are accepted again.
• Checkout requests with redirect URLs -- the success and cancel URLs on a checkout request could fail signature verification because of a JSON encoding mismatch between the app and the server. Encoding is now deterministic on both sides.

2.14.0

2026-05-05

Added

• In-context help during onboarding (#351) -- every onboarding step now has a discreet "Need help?" row with three chips: GPT, Claude, and Copy prompt. Each builds a context-aware prompt for the current step (and permission status, where relevant) and hands it off to ChatGPT or Claude. The clipboard is always populated, so the prompt is never lost if a deep link fails.
• "Okay, what's next?" walkthrough (#351) -- after the final permission step, a three-screen guide teaches the core action: click into a text field, hold the shortcut, speak, release, watch your words land at the cursor. User-paced with Back, Next, and Replay controls. Respects Reduce Motion.
• Previous navigation in onboarding (#351) -- a Previous button now appears in the coach-mark footer from step 2 onwards, so you can revisit earlier steps without restarting the wizard.
• Onboarding model picker (#349) -- first-run setup now includes a "Choose a transcription model" step. The picker suggests a Whisper model based on your Mac (Apple Silicon with 16+ GB RAM and free disk gets Medium, 8+ GB gets Small, smaller machines get Tiny) and lets you opt out to "Apple Speech only". Nothing downloads until you tap Continue.
• Apple Speech fallback -- when Whisper is unavailable, mid-download, or fails to load, the app silently transcribes via Apple Speech and shows a clear status line. Your saved engine choice is never overwritten on a Whisper failure -- once Whisper recovers, it takes over again.

Changed

• Onboarding layout polish (#351) -- progress dots moved to the top of each coach mark for better hierarchy. Permission steps show the status pill in the header next to the title.
• Settings copy -- the Speech engine card in Settings now states the fallback contract: "Apple Speech is used as a fallback while a model downloads or if it cannot run."

2.13.2

2026-05-01

Changed

• Install flow -- VoxChimp no longer offers to copy itself into Applications. The download disk image is the canonical install path: drag VoxChimp Pro into Applications from there, like any other Mac app. If you launch VoxChimp from outside Applications, you will see a friendly reminder asking you to reinstall from the DMG.
• Auto-updates require Applications -- if VoxChimp is running from outside your Applications folder (Downloads, Desktop, an external drive, or a disk image), automatic updates are paused and the manual "Check for Updates..." button explains why. Move VoxChimp into Applications to re-enable updates. This avoids silent update failures Sparkle would have produced from outside Applications.

Fixed

• Duplicate VoxChimp window after install (#336) -- the in-app move flow could spawn a second instance and leave the original running. Removed the in-app mover entirely (see Changed above), so this can no longer happen.
• DMG layout -- the download disk image now uses the standard drag-to-Applications layout with an Applications shortcut, instead of a flat file list. Branded background image will follow in a later release.

2.13.1

2026-05-01

Added

• Install prompt -- when VoxChimp launches from Downloads or a mounted disk image, a friendly dialog now offers to move the app into Applications. Running from Applications keeps Sparkle updates and macOS permissions stable across launches.
• DMG install hint -- the download disk image now shows a "Drag VoxChimp into Applications to install" arrow, so first-time users know where the app is meant to live.

2.13.0

2026-04-30

Added

• First Run Setup wizard -- a guided 8-step welcome tour on first launch covers dictation, the shortcut, vocabulary, AI tools, history, and access permissions. Replay any time from Help > Show First Run Setup or Settings > General.
• Permissions walkthrough -- microphone, accessibility, and input monitoring each have a dedicated card explaining why VoxChimp needs them, with step-by-step instructions and a "Test your hotkey" button to verify Input Monitoring works.
• Live permissions hub -- Settings > Permissions now shows real-time status for each permission with one-click deep links into System Settings. Replaces the previous read-only placeholder.
• Troubleshooting article at voxchimp.com/support#common-issues covers the most common permission and hotkey problems.

Changed

• Chimps tab marked "Coming Soon" with a clearer badge. The standalone custom-dictionary input was removed -- Workspace > Vocabulary already covers custom words.

Fixed

• Microphone permission denial now shows a clear actionable error before recording starts instead of silently failing.
• First-launch onboarding is reliably visible again. Earlier versions silently dropped the wizard after a sidebar refactor.

2.12.1

2026-04-26

Changed

• Internal security infrastructure -- the app now signs every request to the VoxChimp backend with a per-request HMAC. Groundwork for stricter authentication that will roll out in later releases. No user-visible behaviour change in this version.

2.12.0

2026-04-25

Added

• Custom vocabulary -- teach VoxChimp the names, jargon, and acronyms it keeps mishearing, with a guided pronunciation capture flow. Works across WhisperKit, Parakeet, and Apple Speech
• Editable transcripts in History -- correct typos and phrasing in past transcriptions; the app suggests adding repeated corrections to your vocabulary so it learns over time

Removed

• "Reset all settings" option removed -- individual setting changes remain reversible

2.10.6

2026-04-22

Fixed

• Recording widget no longer shows the previous cycle's state -- starting a dictation after a skill run now correctly resets the bar to "Listening"
• Agent and search dashboards stripped of fabricated sample metrics; both views now display only real provider data (connection status, configuration)

2.10.5

2026-04-21

Fixed

• Accessibility permission cycle -- text insertion no longer triggers repeated permission prompts when Accessibility access is revoked mid-session; the app detects the loss, blocks paste gracefully, and notifies the user instead of looping

2.10.4

2026-04-20

Fixed

• Large Whisper model CoreML compile failures now fall back from ANE to CPU automatically, with a persistent banner when the device cannot load the large model
• Transcription view no longer hangs on app launch when scanning the models directory
• Skill hotkey release no longer triggers duplicate invocations
• WhisperKit concurrent transcription crash resolved
• Licence OTP send failures now surface a clear retry message instead of silently failing

Changed

• API dependencies patched for security and maintenance updates

2.10.3

2026-04-18

Fixed

• Translation no longer reverts to the wrong language after switching providers
• License activation dialog hierarchy and keyboard shortcut mappings corrected
• Web search errors are now sanitised and surface useful messages instead of raw provider output
• Logging cleanup across services (replaced stray `print` calls)

2.10.2

2026-04-17

Added

• Changelog extraction for Sparkle appcast release notes
• Sidebar keyboard shortcuts displayed next to navigation items
• History search now shows human-readable dates

Fixed

• Sidebar alignment -- WORKSPACE and APP section labels now align with content headings
• ModelManager test hang resolved with lazy scanning and dependency injection
• UK English spelling enforced across all user-facing copy
• Code coverage and tech debt improvements

Changed

• Website refreshed with updated styling, animations, and layout

2.10.0

2026-04-15

Fixed

• Audio archiving now saves recordings correctly to ~/.voxchimp_audio/

2.9.0

2026-04-14

Added

• Parakeet TDT transcription engine as an alternative to WhisperKit - faster and more accurate on Apple Silicon
• Dark theme across the entire app with a modern look
• Sidebar navigation in settings for easier access to all options
• Card-based settings layout for a cleaner, more organised interface

Changed

• Settings redesigned with sidebar navigation replacing the old tab layout
• Menu items reorganised for better discoverability

Fixed

• Menu changes for better discoverability and usability

2.8.5

2026-04-08

Fixed

• AI translation works more reliably across different language pairs

2.8.4

2026-04-08

Added

• Output locale setting to control spelling normalisation separately from recognition language
• Recognition language and output locale are now independent - recognise in one language, spell in another

2.8.3

2026-04-08

Fixed

• Double window no longer appears on launch
• UI text cleaned up to use consistent formatting throughout
• Licence session handling hardened for better reliability

2.8.2

2026-04-07

Added

• Account protection against fraudulent chargebacks - your licence is better protected

Fixed

• Stability improvements across the licence system

2.8.1

2026-04-05

Fixed

• Stability improvements and internal fixes

2.8.0

2026-04-04

Added

• Recommended models page to help you pick the best local transcription model
• Model search to find and download models directly from the app

Fixed

• Security improvement in API request handling

2.7.1

2026-04-02

Added

• Serper search provider for web-connected skill responses
• Automated release pipeline for faster, more reliable updates

Fixed

• Update dialogue no longer shows duplicate signature warnings

2.7.0

2026-04-01

Fixed

• Billing no longer creates duplicate subscriptions in rare cases

2.6.2

2026-03-30

Fixed

• Build reliability improvements for consistent updates
• Dependency upgrades for better performance and security

2.6.1

2026-03-28

Added

• LM Studio support as an AI provider for skills - run local language models on your Mac or network

Changed

• LM Studio URL is configurable for accessing models on other machines

2.6.0

2026-03-26

Added

• Microsoft Copilot as an AI provider for skills
• Improved Gemini support with better error handling and fallbacks

2.5.1

2026-03-25

Fixed

• AI provider validation works correctly for all supported providers
• Gemini provider connection issues resolved
• Main window opens reliably from the menu bar

2.5.0

2026-03-24

Added

• Multi-provider AI support - choose between different AI services for skills

Fixed

• Microphone permissions checked before requesting access to avoid unnecessary prompts

2.4.5

2026-03-23

Fixed

• Stability improvements and internal fixes

2.4.4

2026-03-23

Fixed

• Menus reorganised for better usability
• Number formatting now respects your system locale

2.4.3

2026-03-23

Fixed

• Stability improvements and internal fixes

2.4.2

2026-03-23

Fixed

• Stability improvements and internal fixes

2.4.1

2026-03-22

Fixed

• Keychain sync prevented from syncing licence data via iCloud to other devices
• Analytics no longer sends events in debug builds

2.4.0

2026-03-19

Added

• Transcription insights dashboard showing your usage patterns and statistics
• Onboarding wizard guides you through setup on first launch

Fixed

• Skill output now returns only the result text without extra preamble

2.3.0

2026-03-15

Added

• Skill system with 5 built-in skills - SOAP notes, meeting summaries, email drafts, rewriting, and grammar fixes
• Hold Right Option to activate skill mode for AI-powered text processing

Fixed

• Purchase button no longer creates duplicate checkout sessions when tapped quickly

2.2.2

2026-03-10

Changed

• Menu bar icon now shows an animated waveform while recording

2.2.0

2026-03-06

Added

• Improved trial security with device fingerprinting to prevent abuse
• Better account protection against licence key sharing

2.1.1

2026-03-04

Fixed

• Dependency updates for improved security and stability

2.1.0

2026-03-02

Fixed

• Stability improvements and internal fixes

2.0.5

2026-02-24

Fixed

• Code quality improvements across the app for better reliability
• Linting fixes throughout the codebase

2.0.4

2026-02-16

Fixed

• Window reopen now works reliably from the menu bar after closing

2.0.3

2026-02-16

Fixed

• Menu bar visibility toggle and window reopen from menu bar

2.0.2

2026-02-16

Fixed

• Auto-updater signing fixed so updates install correctly

2.0.0

2026-02-01

Added

• Post-transcription translation using Apple Translation framework
• WhisperKit integration for on-device transcription with model selection
• Apple Speech fallback when Whisper model is not yet downloaded
• Menu bar status item with chimp mascot for quick access
• Model loading indicator with progress feedback during downloads
• Voice health dashboard with real-time acoustic analysis
• Health notifications for voice biomarker alerts
• Audio level indicator showing microphone input in real-time
• Licensing system supporting App Store and direct distribution builds
• Sparkle update framework for automatic updates

Changed

• Complete UI redesign with warm theme and modern aesthetics
• New app icon featuring the chimp mascot
• Modernised interface with vibrancy effects and SF Symbols
• Improved error recovery with graceful degradation on failures

Fixed

• Race conditions in model download flow
• WhisperKit model naming and cache scanning issues
• Toolbar button functionality
• Health window crash on launch
• Baseline accumulation bug in health analysis

1.0.0

2026-01-15

Added

• Initial release of VoxChimp voice dictation
• Hold-to-record with configurable hotkey (Option, Control, Caps Lock)
• Local Whisper transcription (tiny/small/medium/large models)
• Auto-paste transcribed text at cursor position
• Basic grammar fixing (capitalisation, punctuation)
• Transcription history with JSON export
• Optional audio archiving
• Native macOS UI
• Comprehensive logging with rotation