Changelog

TypeWhisper 1.2.2 lets you dictate into the World of Warcraft chat and makes pasting into Flutter apps work. The 1.2.0 release notes describe what is new in 1.2, including managed settings for organizations and the machine-wide MSI.

Dictate into the World of Warcraft chat

• Turn on World of Warcraft under Settings › Advanced › Integrations and click Install add-on. TypeWhisper finds the game through Battle.net and installs the TypeWhisper add-on into every version on the PC: Retail, Classic and Forever.
• While the game is in front, your dictation shortcut opens the chat, inserts the text and sends it. A chat that is already open stays open with its text. Long text goes out as several messages of up to 255 bytes, and a message never starts with a slash, so a dictation cannot run a game command.
• The add-on shows the dictation as a cast bar in the game's style: a waveform while you speak, and a bar that fills while TypeWhisper transcribes.
• Sending, the cast bar and an optional translation of your messages are set on the add-on's page in the game: type /tw. Translation uses the default LLM of your workflows. TypeWhisper reads these settings when you type /reload or log out.
• The add-on is also available on its own from TypeWhisper/typewhisper-wow.

Fixes

• Pasting into apps built with Flutter works. TypeWhisper now sends the keys of a paste, of the Enter after it and of selected-text capture with the scan codes a real keyboard sends, so Flutter no longer receives a plain "v" instead of Ctrl+V. (#732)

Updating

• 1.2.0 and 1.2.1 installations are offered 1.2.2 through the update notice. Download and restart installs it and keeps the profile, plugins and settings.
• Installations of TypeWhisper 1.1 and 1.0 update to 1.2.2 directly, as described in the 1.2.0 release notes.
• Machine-wide installations are updated by deploying the 1.2.2 MSI.
• The Microsoft Store version is updated separately.

Installation and requirements

• The per-user installer, the MSI and the portable ZIP are available for x64. Setup installs the .NET 10 runtime when it is missing; the portable ZIP requires it separately. The Group Policy templates are attached as TypeWhisper-win-PolicyTemplates.zip.
• TypeWhisper requires Windows 10 build 19041 or later.
• As with 1.2.0, the GitHub installers, including the MSI, are not code-signed, so Windows SmartScreen may ask for confirmation.

TypeWhisper 1.8.2 is an update to the stable 1.8 macOS release. It makes ElevenLabs dictations much faster, stops cloud engines from uploading audio again for the live preview, lets you import your own Parakeet models, adds word timestamps to the HTTP API and fixes dictation into Flutter apps and the kitty terminal.

Highlights

• ElevenLabs 1.1.0: realtime dictations in Auto mode no longer wait about 15 seconds after the transcript arrives. In our tests a 3.5-second dictation took about half a second instead of 16 seconds. The plugin also adds the Scribe v2 Medical model, more languages and word timestamps.
• Live preview: with a cloud engine that cannot stream, such as Microsoft AI, Groq or ElevenLabs, the live preview no longer uploads the last 10 seconds of audio again every 3 seconds. It runs on a loaded local engine instead, while the final text still comes from your dictation engine.
• Parakeet 1.7.0: Import Model… adds your own Parakeet TDT models converted to Core ML, for example a fine-tune for one language.
• HTTP API: verbose_json responses can include the time of each word.

Dictation

• With Live preview engine set to Match dictation engine, the preview now runs on a loaded local engine that streams your language, such as Apple Speech, while the final text still comes from your dictation engine. Without such an engine no preview is shown, unless live text in the text field or the Live Transcript window is on; these keep using the cloud engine then. To keep the previous behavior, choose your cloud engine under Settings > Appearance > Live preview engine. Before, the preview sent about three times the audio of each dictation to the cloud engine.
• Dictating into Flutter apps for macOS shows the text in the field. TypeWhisper wrote it into Flutter's hidden input view, which reported success while the visible field stayed empty; it now pastes there.
• Dictating into the kitty terminal no longer ends with "No text field detected" when kitty shows the text a moment late. kitty passes typed text on to the shell, which echoes it back, so a field that has not changed yet does not mean the text went nowhere. TypeWhisper now counts the dictation as inserted there, as it already does for web inputs.
• Apple Speech offers Apple's regional Chinese models separately, so you can pick Hong Kong Cantonese with Traditional Chinese characters. Before, Chinese could fall back to the model of another region.
• The "New in TypeWhisper" window has a single Continue button and stays in front of Settings while TypeWhisper is active.

Plugins, API and CLI

• ElevenLabs 1.1.0: realtime dictations in Auto mode end as soon as the transcript arrives; the plugin used to wait for ElevenLabs to close the finished session. Auto mode still uses the batch API for Scribe v2 Medical, with active dictionary terms and for recordings over 30 seconds, so those take as long as before. If a realtime session sends no final transcript, the batch API transcribes the whole recording instead of returning partial text.
  • Transcriptions through the batch API keep Scribe's word timestamps. verbose_json API responses get segments split at sentence ends and pauses instead of an empty segments list, and word timestamps when requested. Speaker detection aligns ElevenLabs transcripts by word as well. Short realtime dictations report word timestamps and, with automatic language detection, the detected language.
  • The new Scribe v2 Medical model, tuned for clinical audio at the Scribe v2 price, transcribes in batch. A saved selection of the deprecated Scribe v1 moves to Scribe v2.
  • The detected language is reported with the two-letter code TypeWhisper uses (de instead of deu), so German and other language-specific formatting applies with automatic detection.
  • Javanese, Asturian, Cebuano, Irish, Igbo, Kabuverdianu, Kurdish, Kyrgyz, Ganda, Luo, Northern Sotho, Odia, Umbundu, Wolof, Xhosa and Zulu can be selected; Hawaiian and Volapük, which ElevenLabs rejects, are no longer offered.
  • Realtime errors whose type does not say "error", such as rate_limited, quota_exceeded or invalid_request, switch to the batch API at once instead of after an eight-second wait. When an API key's own credit quota is used up, the error shows ElevenLabs' message instead of "Invalid API key".
• Parakeet 1.7.0: Import Model… in the plugin settings adds your own Parakeet TDT models converted to Core ML, from a folder or a Hugging Face repository. Raw .mlpackage conversions are compiled during the import. Fine-tunes that add tokens to the vocabulary, such as the Greek Parakeet v3 fine-tune, transcribe correctly. Imported models appear in the model list and can be deleted there.
• Language Detection 1.0.1 tells Hong Kong Cantonese (zh-HK) apart from Mandarin. Before, all Chinese variants counted as one language, so dictation always used the first Chinese language you selected.
• HTTP API: verbose_json responses can include the time of each word. Send timestamp_granularities[]=word (as in OpenAI's API) to get a words array with word, start and end in seconds; /v1/transcribe/local-file takes "timestamp_granularities": ["word"]. Responses without the parameter stay unchanged. Engines that report word times, such as Parakeet and ElevenLabs, fill it.
• CLI: typewhisper completions bash|zsh|fish prints a shell completion script. It completes commands and options, file paths, and engine and model IDs from the running app. The CLI is part of the direct download edition.

Compatibility and distribution

• Requires macOS 14.0 or later and supports Apple silicon and Intel Macs.
• Every direct download release includes the iCloud bridge. The Mac App Store edition ships 1.8.2 from the same commit.
• Parakeet 1.7.0 requires TypeWhisper 1.8.2. ElevenLabs 1.1.0 runs on TypeWhisper 1.8.0 and later; word timestamps through the HTTP API need 1.8.2. Language Detection 1.0.1 requires TypeWhisper 1.8.1.
• Built with Xcode 27.1.

TypeWhisper 1.8.1 is an update to the stable 1.8 macOS release. It adds the Language Detection plugin, a quick way to correct the last dictation, IBM's Granite Speech 5.0 models and the Local LLM (llama.cpp) plugin, and makes the built-in workflows follow your dictation more reliably. It is also the first update of the Mac App Store edition.

Highlights

• Language Detection (new plugin): when you select several spoken languages and the engine accepts only one, such as Parakeet, WhisperKit, Qwen3 or Groq, the plugin detects which of them you speak instead of always using the first one. It runs on your Mac in about 15 ms per transcription, and the audio never leaves it. Canary, Cohere Transcribe (Local) and Google Cloud STT cannot detect the language themselves; with the plugin they now work with Auto-detect all languages. Install it from Discover plugins; the Mac App Store edition includes it. Live previews still use the first selected language.
• Correct Last Dictation: a menu bar action and an assignable shortcut open a small editor with your latest dictation. Saving updates the History entry, copies the corrected text and learns a single replaced word for the Dictionary with the existing conservative rules.
• Toggle dictations can stop on their own after a chosen silence (Dictation > Toggle Mode, off by default, 3 seconds). They never stop before you start speaking.
• The dictation indicator can show the current model and switch it from a menu (Appearance > Left Side or Right Side > Model). A model chosen during a recording applies to the next dictation.
• LLM workflows can use the text of the front window as reference, for example to spell names as they appear on screen ("Use screen context", off by default, needs Screen Recording permission).
• The workflow editor shows Behavior and Trigger first and moves everything else into Advanced. A new Try It Out card runs the unsaved draft.
• Granite Speech adds IBM's Granite Speech 5.0 models. Granite 5.0 TurboCTC (470M, about 950 MB) is the new default: it transcribes English, French, German, Spanish, Portuguese, Italian and Japanese with punctuation and capitalization in a few milliseconds per sentence, but cannot translate or use dictionary terms. Granite 5.0 2B (about 4.6 GB) also writes punctuated text, translates into a target language you choose and uses your dictionary terms. If you already use a Granite 4.0 model, it stays selected.
• Local LLM (llama.cpp) (new plugin): runs GGUF models on your Mac with llama.cpp and Metal for workflows and prompts, without an API key. It offers the same Gemma 4, Qwen3.5 and LFM2.5 models as the Windows plugin, and you can add any single-file GGUF model from Hugging Face or import a .gguf file. Install it from Discover plugins; it needs a Mac with Apple silicon.
• Spanish is available as an app language.

Dictation

• English spoken punctuation accepts US terms ("exclamation point", "open square bracket", "close square bracket"), spoken quotes ("open quote", "close quote") and hashtags ("hashtag" becomes #).
• Normalize spoken numbers to digits supports Polish: dziesięć becomes 10, dwadzieścia trzy 23 and dwa przecinek pięć 2,5. Declined forms and a lone jeden stay as spoken.
• With several spoken languages, a language without number parsing no longer lets a later language rewrite its words. With Polish listed before English, Polish ten ("this") and one ("they") no longer become 10 and 1.
• Inline Commands: a spoken instruction such as "translate this into English" now changes the output language instead of keeping the language of the dictation.
• Dictating into GNU Emacs, a remote desktop client such as Parsec, or a new message in Apple Mail no longer ends with "No text field detected". These apps either report only their window to Accessibility or do not show their text changing, so TypeWhisper no longer takes that as a missed text field. Web pages without a text field still get the warning.
• With Preserve clipboard content on, Telegram and other apps that report Accessibility writes as failed no longer receive a dictation twice.
• Chinese, Japanese and other languages written without spaces are counted by characters for the "Skip AI post-processing" word threshold, History and Statistics. Until now, Chinese dictations always skipped AI processing, no matter how long they were.
• The setup wizard recommends a local engine that supports your spoken language instead of always Parakeet.
• Pickers that fall back to the default engine name it and link to the Dictation page.
• Parakeet Vocabulary Boosting no longer replaces ordinary words with dictionary terms ("my grandpa" no longer becomes "my Grafana").

Workflows and language models

• The built-in workflow instructions follow your dictation more reliably on local and cloud models. Cleaned Text also removes filler words, stutters and repeated words, keeps only the final version when you correct yourself, and splits run-on speech into sentences. Translation drops filler words too. Email Reply writes directly to the person you name and ends with a closing instead of a placeholder. Meeting Notes stay in the language of the dictation and keep every date. A workflow's fine-tuning still takes priority.
• New installs get a Translate preset that reliably translates German text and a Draft Email preset that no longer invents details. Prompts you already have keep their text.
• An Always workflow can exclude apps and websites.
• History details show the LLM provider, model and the tokens the provider reported for each processed entry. /v1/history returns them as llm_usage.
• Local LLM (MLX) recommends a model for this Mac based on its memory, describes each Gemma 4 model and warns when a model needs more memory than the Mac has. Gemma 4 E4B (8-bit) is no longer experimental.

Recorder and meetings

• Settings > Recorder > Recordings Folder sets where new recordings are saved, for example a folder outside iCloud Drive. If the folder is unavailable, a recording goes to the default folder and the Recorder says so.
• After Stop, the recording is transcribed in the background, so the next recording can start right away.
• Meeting automation detects Google Meet calls at meet.google.com/meet/<code> addresses again.
• Meeting automation can run without countdowns and the indicator ("Show countdowns and indicator for meeting recordings").
• Deleting a recording asks first and moves the audio and its sidecar files to the Trash instead of deleting them for good.
• Translate is offered only for models that can translate. A Translate choice made for an earlier model no longer makes every file fail after you switch to a model that cannot translate.
• Granite's Translate lets you choose the target language in the Recorder, file transcription and dictation recovery: English, French, German, Spanish, Portuguese, Japanese, Italian or Chinese. English is the default. Until now Granite translated into the selected spoken language.
• Obsidian 1.2.0 can export Recorder transcripts as notes (Auto-Export > Recorder transcripts, off by default).
• Recorder completions for webhooks, Script Runner and GET /v1/recorder/recordings include speakers, the recording time, the duration and the calendar event.
• Speaker detection keeps your own speech that the microphone detection missed with "Me". When you rename your microphone speaker, later recordings use that name. Recordings transcribed with engines that return no timestamps, such as Qwen3, get speakers through a Parakeet timing pass.

Dictionary

• Import your own term packs from a JSON file in the community pack format under Dictionary > Term Packs > Local Packs, turn them on and off like the other packs, and export them for another Mac. Local packs stay on this Mac: TypeWhisper does not review, sync or update them.

Plugins and SDK

• New: Language Detection 1.0.0 and Local LLM (llama.cpp) 1.0.0.
• Parakeet 1.6.1: using another Parakeet model for a single transcription, as the local API's model, the CLI's --model or a workflow with its own Parakeet model do, now waits for that model to load. Until now it failed with "Cloud provider not configured. Please set an API key." Nemotron 3.5 ASR Streaming no longer drops the first and last words of files and other recordings it transcribes after they end.
• Granite Speech 1.2.0 (see Highlights). Translation also works while dictionary terms are active; the terms are now passed as keywords after the instruction, as the model expects. The local API's target_language with task=translate goes straight to Granite when it supports that language.
• Canary 1.1.1 and Cohere Transcribe (Local) 1.1.0 no longer stop with an error under Auto-detect all languages when nothing picks the language. They use the first of your preferred languages in System Settings > General > Language & Region that they support. Google Cloud STT 1.1.1 keeps using its default language code in that case.
• Canary 1.1.1, Qwen3 1.1.15, Voxtral 1.0.20 and Local LLM (MLX) 1.0.7: each of these plugins brings its own copy of Apple's MLX framework. With several of them enabled, one plugin could run code from another plugin's copy, which can crash TypeWhisper during a transcription or text processing. Each plugin now uses only its own copy, as Granite Speech already does.
• R2T2 offers only the own-server mode on Intel Macs, where the built-in server cannot run.
• MCP Client: an action run right after saving a server keeps its connection, and cancelling during a connection is reported as cancellation.
• Plugin SDK: optional LLM usage reporting (PluginLLMUsageReporting), spoken language identification providers (SpokenLanguageIdentificationProviderPlugin) and engines that leave language detection to the host (HostLanguageDetectionEnginePlugin). Engines can report translation support, translation target languages and dictionary term support per model (supportsTranslationForModelId:, translationTargetLanguagesForModelId:, supportsDictionaryTermsForModelId:).

Mac App Store edition

• GigaAM (local Russian dictation) and Language Detection are included.
• The Premium page offers Connect Existing TypeWhisper License for customers who bought a license outside the App Store.
• The edition keeps access to a chosen Recordings Folder after a restart.

Fixes

• Closing the last window no longer quits TypeWhisper on macOS 15; the app keeps running in the menu bar.
• Pair Device stays available after launch when paired devices are allowed.
• The German interface translates the API token setting and the Script Runner rules.

Compatibility and distribution

• Requires macOS 14.0 or later and supports Apple silicon and Intel Macs.
• Every direct download release includes the iCloud bridge. The Mac App Store edition ships 1.8.1 from the same commit.
• Parakeet 1.6.1, Canary 1.1.1, Language Detection 1.0.0, Cohere Transcribe (Local) 1.1.0 and Google Cloud STT 1.1.1 require TypeWhisper 1.8.1. Granite Speech 1.2.0, Qwen3 1.1.15, Voxtral 1.0.20, Local LLM (MLX) 1.0.7 and Local LLM (llama.cpp) 1.0.0 also run on TypeWhisper 1.8.0; the target language choice for Granite needs 1.8.1.
• Built with Xcode 27.1.

TypeWhisper 1.8.0 is the stable 1.8 macOS release. It adds Premium speaker detection, lets paired iPhones transcribe on the Mac, brings managed settings for organizations, and is the first release on the Mac App Store.

Highlights

• Premium speaker detection shows who said what. Detection runs on the Mac. History plays recordings with a speaker-colored scrubber and can show the transcript as a conversation. Speakers can be renamed, merged, split, and reassigned, with undo. A new Speakers page in Settings transcribes dropped files with speakers and lists the people TypeWhisper knows by voice.
• iPhone and iPad (Premium): a paired iPhone running TypeWhisper for iOS 1.2 can transcribe with this Mac's models over the local network. With "Allow Writing with This Mac" turned on, it can also use the Mac's language models and workflows. Both are off by default. Pairing uses a six-digit code shown on both devices, and the connection is encrypted.
• Managed settings: administrators can configure TypeWhisper through macOS configuration profiles. They can restrict cloud transcription and text processing, limit providers and plugins, block plugin and model downloads, enforce a transcription or language model, and control updates, launch at login, setup, and optional prompts.
• TypeWhisper is now available on the Mac App Store. That edition ships the same version from the same commit as the direct download.
• History entries with saved audio can be re-transcribed with another engine, model, language, or workflow. The new result is shown next to the current text, and you decide which version to keep.
• The menu bar has quick selectors for microphone, spoken language, and model.
• Recorder transcriptions skip stretches without speech, so a quiet start to a call no longer produces invented text.
• The local API now requires the API token by default. See Compatibility and distribution below.

Dictation and shortcuts

• Opt-in microphone pre-roll keeps the last 500 ms of audio in memory, so words spoken right at the hotkey press are not cut off. It applies to non-Bluetooth inputs, and the macOS microphone indicator stays on while it is enabled.
• Direct download edition: when a dictation ends without a text field to paste into, the indicator shows "No text field detected" with an Insert button. Click into the field you want, then click Insert or press your Paste Last Transcription shortcut. The offer stays for 30 seconds, and hovering pauses it. The message for a text field that changed during dictation offers the same button.
• Chromium browsers, Electron apps, and other web content no longer trigger false "No text field detected" warnings. Each indicator message can be dismissed individually.
• The Paste Last Transcription shortcut now pastes when it uses the V key. When the focused field stays unchanged, it says "Text may not have been inserted" instead of "Text inserted".
• When a live transcription connection drops or stops returning text, TypeWhisper transcribes the full recording instead of inserting the partial preview.
• A cloud dictation no longer waits a minute on "Transcribing…" when the live preview was still sending audio at stop.
• If the selected engine is briefly unavailable, for example while a plugin updates, TypeWhisper uses another engine only until the selected one is back. Deleting the last model of a local engine keeps its plugin enabled and the engine selected. The Recorder and file transcription keep their engine through plugin updates.
• An auto-unloaded local model starts loading when recording starts, and the indicator shows "Loading model…" while it loads. Loaded Parakeet models and Apple Intelligence workflows are prewarmed at recording start.
• A dictation that contains only digital silence ends with a hint to check the mute or reconnect the headset. A Bluetooth input that starts silent is rebuilt once, and the indicator warns "No sound from the microphone. Is it muted?" during the recording.
• Audio ducking restores the output volume from before microphone startup, including on USB speakerphones and after cancellation.
• The menu bar shows the dictation shortcut, labels the meeting Recorder as "Start Recorder", and keeps the last dictation error with a "Fix in Settings…" entry. Errors appear even when the indicator is set to Never.
• With "Preserve clipboard content" off, dictations appear in clipboard history apps.
• Fixed double insertion in cmux, a leading space and lowercase first letter in Firefox composers, spaces before Japanese full-width punctuation, and stray spaces next to the kana prolonged sound mark.
• Typing and clicking no longer freeze system-wide when TypeWhisper stops responding. Also shipped in 1.7.1.

Recorder and meetings

• Recorder transcriptions send only the parts with speech to the engine, found with the sound classifier built into macOS. Times in the transcript and in speaker detection still match the recording. A recording without speech says so instead of returning a transcript.
• In a Recorder recording with microphone and system audio, speech from the microphone becomes the speaker "Me".
• Hovering over the recording indicator turns the red dot into a Stop button for the Recorder.
• Meeting automation can detect calls without a calendar event: Zoom and Teams microphone activity and recognized meeting links in the browser. It is off by default.
• Kontur Talk calls are detected in the app and in the browser. New Teams calls are detected through its helper processes.
• "Ask before recording events without other attendees" keeps automatic recording for meetings with other people and asks first for events nobody else is invited to. It is off by default.
• Saved Recorder transcripts can be sent to Webhook Notifications or Script Runner, with a separate, default-off option for each. Webhook and Script Runner 1.2.0 require TypeWhisper 1.8.0.
• WhisperKit retries failed audio chunks instead of returning a silently incomplete transcript.

Text handling and workflows

• Dictionary corrections can use regular expressions. Existing entries stay literal. Regex rules are kept in backups, sync, and the HTTP API, and syncing them requires the current Mac and Windows apps.
• Term pack entries use Precise Vocabulary Boosting, so words like "possible" or "notify" no longer turn into Ansible or Netlify. Changes to pack entries survive turning packs off and on and pack updates.
• When you add a Term while Parakeet is selected and Vocabulary Boosting is off, TypeWhisper offers to turn it on, including the one-time model download (about 100 MB).
• Learn from Corrections can wait until a correction was made 2, 3, or 5 times within 30 days. Changing one number into another is no longer learned as a correction.
• The Filler Words plugin removes "um", "ah", and "eh" only from English text, so "um 10 Uhr" keeps its preposition.
• An LLM reply that stopped at the output token limit counts as a failure instead of being inserted as partial text. A fallback list moves on to the next provider.

History, files, and sync

• Re-transcribe compares results side by side and keeps the original. Cloud providers are named, with a note that usage charges may apply, before anything is sent.
• File Transcription, watch folders, and the local API accept Opus files, such as WhatsApp voice messages, where macOS can decode them.
• Importing a backup onto the Mac it came from no longer duplicates history, workflows, profiles, or prompt actions. "Replace existing items" (typewhisper import --replace, mode=replace in the local API) overwrites items with the same name.
• Recovery lists every recording with its own Discard button. Preserved recovery recordings are stored as AAC instead of WAV.
• Sync settings list synced devices and let you remove them. Deletions now stick in Automatic iCloud Sync.

Settings, licensing, and administration

• Installed plugins have their own sidebar pages. Integrations becomes Discover plugins, and plugin updates appear in the plugin page header.
• A short, skippable "New in TypeWhisper" overview introduces selected features once per update on a normal launch. Dictation and other work take priority. It can be reopened from Settings > About.
• Settings > License follows the pricing page of the website. Individual and Team are sold monthly or as Lifetime. Enterprise is a subscription for up to 50 devices at 199 EUR per month or 1,990 EUR per year. It is no longer sold as Lifetime, and existing Enterprise Lifetime licenses keep working.
• Managed settings cover 18 keys. Settings > Advanced > Managed Settings shows what the organization manages, and locked controls say "Managed by your organization." See the managed settings guide. Example profiles, a profile generator, and Jamf Pro and ProfileManifests schemas are in the repository.

Local API, CLI, and automation

• typewhisper models load, unload, and delete manage local models from the command line. typewhisper transcribe - reads audio from stdin.
• GET and PATCH /v1/settings/audio read and change microphone priority, ducking, media pause, and sound feedback.
• New authenticated routes name speakers, enroll voices, and confirm voice profile suggestions. History entries return speaker segments, suggested names, and the Recorder recording_id. Transcription requests accept detect_speakers and speaker_count.
• GET /v1/recorder/recordings returns completed Recorder transcripts, also after a restart.
• /v1/llm/models, /v1/llm/process, /v1/workflows, and /v1/workflows/run process text with this Mac's language models and workflows without inserting it.
• Browser extensions can report the active tab through PUT /v1/browser/context, which lets website workflows match in browsers without AppleScript support and in the Mac App Store edition.
• verbose_json responses apply dictionary corrections to segment texts.

Models and integrations

• Parakeet Ultra is the default Parakeet model for new selections. Saved selections stay. Parakeet no longer writes a literal <unk>.
• Fluid Speech Mini joins the Parakeet plugin as a small English model with a 250 MB download.
• The new GigaAM plugin provides local Russian dictation with punctuation and capitalization. It runs on the CPU, downloads a 230 MB model on first use, and requires TypeWhisper 1.8.0.
• Cloud transcription plugins split long recordings into chunks below their provider's limits and wait longer for long jobs. These plugin releases require TypeWhisper 1.8.0; 1.7 keeps its compatible releases.
• Gemini offers Verbatim (the new default) and Smart transcription. Gemini Live finishes sooner after the hotkey is released, and a crash between Gemini Live dictations is fixed.
• The Claude plugin reads replies that start with a thinking block and adds reasoning effort, prompt caching, and clearer error messages.
• OpenAI Compatible profiles can upload WAV directly, and Batch profiles no longer upload audio for a live preview.
• Soniox sends its API key in the WebSocket authorization header ahead of Soniox's authentication change.
• Plain HTTP to Tailscale addresses and MagicDNS names is allowed, for example for LM Studio.
• Parakeet shows download progress, also in the setup wizard, and a model switch no longer waits for the previous model's boosting model.
• Fixed a crash after uninstalling a plugin and a crash in Local LLM settings while a model loads.
• Wrong and missing translations were fixed, and Japanese plugin wording was improved.

Accessibility

• The app and the bundled plugins' settings are labeled for VoiceOver. Buttons name their action and target, and status that was only shown by color, icon, or tooltip is spoken.
• Indicator feedback, recording start and stop, Recorder events, and errors are announced while VoiceOver runs.
• While recording, the menu bar icon becomes a filled badge, so the state no longer depends on color. A rejected undo during a recording is shown on the indicator, and copy, delete, and export actions confirm on screen.
• LLM fallback rows can be reordered with buttons. Dictionary, snippet, and file transcription cards open with the keyboard, and the workflow editor saves with Command-S and closes with Escape.
• With Reduce Motion, the recording dot no longer pulses and the indicator no longer pops in.

Compatibility and distribution

• Requires macOS 14.0 or later and supports Apple silicon and Intel Macs.
• Every direct download release includes the iCloud bridge. The Mac App Store edition uses its iCloud container directly.
• The Mac App Store edition ships the same version from the same commit. It is sandboxed, updates through the App Store, sells Premium through in-app purchase, and includes its plugins in the app. Sparkle updates, the CLI, license keys, correction learning, dictation undo, and the "No text field detected" Insert offer are direct download only. Speaker detection in the Mac App Store edition requires Apple silicon.
• The local API server now requires the API token by default. The CLI, the MCP server, and the Pi extension already send it. The Raycast extension needs its next update. A server that runs without a token during the update keeps doing so, and Settings warns until Require API Token is turned on. Settings > Advanced can copy the token. Regardless of the token setting, the server rejects requests that websites send through the browser.
• New plugin releases require TypeWhisper 1.7.0 or later. Plugin releases that use SDK features new in 1.8, such as Parakeet 1.4.0, GigaAM, Webhook and Script Runner 1.2.0, and the long-recording cloud plugins, require TypeWhisper 1.8.0.
• Built with Xcode 27.1.

TypeWhisper 1.2.1 fixes two issues in 1.2.0. The 1.2.0 release notes describe what is new in 1.2, including managed settings for organizations and the machine-wide MSI.

Fixes

• A shortcut without Ctrl, Alt or Shift can again use Insert, Pause or a media, volume, browser or launch key, as in TypeWhisper 1.0. Since 1.1 only F1–F24 worked alone, which made push-to-talk awkward on laptops whose F-keys need Fn. Keys that type or edit text, toggle a lock or are used by Windows itself still need a modifier, and the message now lists the keys that work alone. A single Insert shortcut carried over from 1.0 registers again after the update. (#700)
• The Hugging Face token field of Cohere Transcribe (Local) and Supertonic TTS shows "(optional)" once instead of twice. (#703)

Documentation

• The managed settings guide lists the hosts from which the Granite Speech plugin downloads its Python runtime. AllowModelDownloads = 0 blocks this runtime download together with the model. (#702)

Updating

• 1.2.0 installations are offered 1.2.1 through the update notice. Download and restart installs it and keeps the profile, plugins and settings.
• Installations of TypeWhisper 1.1 and 1.0 update to 1.2.1 directly, as described in the 1.2.0 release notes.
• Machine-wide installations are updated by deploying the 1.2.1 MSI.
• The Microsoft Store version is updated separately.

Installation and requirements

• The per-user installer, the MSI and the portable ZIP are available for x64. Setup installs the .NET 10 runtime when it is missing; the portable ZIP requires it separately. The Group Policy templates are attached as TypeWhisper-win-PolicyTemplates.zip.
• TypeWhisper requires Windows 10 build 19041 or later.
• As with 1.2.0, the GitHub installers, including the MSI, are not code-signed, so Windows SmartScreen may ask for confirmation.

TypeWhisper 1.2.0 brings managed settings for organizations. Machine policies set through Group Policy, Intune or any tool that writes the registry control the license, providers, cloud processing, downloads, updates and startup, and a new MSI installs the app for every user of a device. Installations of TypeWhisper 1.1 update in place.

Managed settings for organizations

• TypeWhisper reads machine policies from HKLM\SOFTWARE\Policies\TypeWhisper (enforced) and its Recommended subkey (defaults users can change). Without policy values nothing changes. The app applies policy changes while it runs.
• ManagedLicenseKey activates a commercial license for each Windows user who starts TypeWhisper, with the same name and behavior as on macOS. While the key is managed, the Premium page hides key entry, deactivation and purchase links.
• Transcription and text models can be enforced or recommended as provider and model pairs. Allowlists limit the transcription providers, text providers and plugins.
• AllowCloudTranscription and AllowCloudTextProcessing keep audio and text on local plugins. Every plugin declares whether it runs locally, in the cloud or depending on its configuration.
• AllowPluginDownloads and AllowModelDownloads stop plugin installs and updates from the catalog and the model downloads the app starts.
• UpdateMode, UpdateChannel and AllowedUpdateChannels control app updates; LaunchAtLogin, SkipOnboardingWhenReady and SuppressOptionalPrompts control startup, setup and the Premium prompt.
• Restrictions apply to dictation, file transcription, workflows, recovery and the HTTP API, which answers 403 forbidden. Pickers, plugin pages and Integrations show why an entry is unavailable.
• Settings > Advanced > Managed settings lists the effective values and any problems. The support report contains the same section.
• Group Policy templates (ADMX with English and German ADML) and the registry script Set-TypeWhisperPolicy.ps1 are attached as TypeWhisper-win-PolicyTemplates.zip. The deployment guide describes every key, Group Policy and Intune, network hosts, the microphone permission and packaging.

Machine-wide installation

• Each release now also ships an MSI, TypeWhisper-win-x64.msi. It installs TypeWhisper once for all users into C:\Program Files\TypeWhisper, with Start menu and desktop shortcuts for all users. Settings, models and plugins stay in each user's profile.
• A machine-wide installation is updated by the deployment: the app does not install app updates itself and locks the update channel. Plugin updates are unaffected.
• A silent MSI installation does not install the .NET 10 runtime. Deploy it first; the guide lists the commands. An interactive installation offers to install it.

Other changes

• After an update, a short overview of new features appears once. It also appears when Settings is opened during the restart, and it never opens during a dictation. Account & about > What's new in TypeWhisper shows it again.
• Select lists with ten or more choices, such as the spoken language, have a search field. It ignores case and accents and also finds language codes.
• The overlay offers the icon of the app you dictate into as its own widget, separate from the app name.
• The tray menu follows the Light theme.
• The Premium page explains what a commercial license adds on Windows, with examples of what learning from corrections fixes. With Premium access it lists the features with their status and settings.
• Clicking a recent transcription on Home opens it in History. Recent rows have Copy and Open in History in their context menu, and the History context menu acts on the row it was opened on.
• Regex corrections in the dictionary keep their own identity through sync with macOS, imports and backups, so patterns such as \d and \D no longer overwrite each other.
• TypeWhisper.exe --page <page>/<setting> opens a settings page scrolled to one setting, for example --page advanced/ManagedPolicy. Search results and links now also reach the settings on the Audio page.
• Failed plugin settings actions are recorded in the diagnostic log, without content.
• Local speech engines in the separate transcription worker no longer fail when they load a component after the worker has been running for a while. (#696)

Updating

• 1.1 installations are offered 1.2.0 through the update notice. Download and restart installs it and keeps the profile, plugins and settings. The app's update component moves to a new version with this release.
• Installations of TypeWhisper 1.0 update to 1.2.0 directly and import their existing data on the first start, as with 1.1.
• The Microsoft Store version is updated separately.

Installation and requirements

• The per-user installer, the MSI and the portable ZIP are available for x64. Setup installs the .NET 10 runtime when it is missing; the portable ZIP requires it separately.
• TypeWhisper requires Windows 10 build 19041 or later.
• As with 1.1, the GitHub installers, including the MSI, are not code-signed, so Windows SmartScreen may ask for confirmation.

TypeWhisper 1.1.0 is the new Windows release. The app is rebuilt on WinUI 3 and follows the layout of TypeWhisper for macOS, and every transcription, text and speech provider is now a portable plugin. Installations of TypeWhisper 1.0 update in place and import their existing data on the first start.

Highlights

• Settings is the main window, with the macOS sidebar, a Home dashboard and Dictionary, Snippets, Workflows, Recorder and File transcription in one place. History opens in its own window. A Theme setting offers System, Light and Dark.
• The interface is available in English, German, Japanese and Simplified Chinese. It follows the Windows display language on the first start and can be changed in Settings.
• Providers are installed on demand from Integrations. Plugin updates appear next to app updates; unless turned off, they are downloaded in the background and applied at the next start.
• Workflows run in matching apps and websites or from their own shortcut. They can set the spoken language and the task, including native translation to English, and dictation shortcuts can also use their own transcription model. A workflow palette runs workflows on the selected text, and "Press Enter after inserting" offers the four macOS modes.
• An optional local HTTP API and a bundled command-line client control the running app. Advanced settings lists the Raycast extension and the browser extension for Chrome and Firefox, which reports the active tab for website workflows.

Dictation

• Escape cancels a recording or its transcription.
• The microphone opens while the model is still loading, so the start of a dictation is kept.
• A new "Paste last transcription" shortcut (unassigned by default) and tray item insert the last dictation again.
• Audio ducking lowers other apps instead of the output volume, so feedback sounds and spoken feedback keep their level.
• Shortcuts sent by gesture and macro tools such as AutoHotkey work, and AltGr characters stay typeable with Ctrl+Alt shortcuts.
• A microphone test and live microphone status explain capture problems. After the Windows default microphone changes, the next recording uses the new one instead of repeating the old error.
• "Paste only into the original field" handles combobox search fields and terminals and falls back to the window when no field is found.

Local models

• Parakeet, Whisper (Local) and Qwen3 ASR run in a separate worker process. A crash ends only that process, and the transcription is retried.
• NVIDIA Parakeet offers Parakeet Ultra 0.6B as the recommended model. On x64 it runs on NVIDIA, AMD and Intel graphics cards through Vulkan.
• Qwen3 ASR (Local) transcribes on the CPU with a 0.6B or 1.7B model.
• Dictionary terms boost Parakeet dictations of up to five minutes and are passed to local Whisper as its prompt.
• A Local LLM plugin runs Gemma 4, Qwen3.5 and LFM2.5. A local model can suggest dictionary aliases.
• Parakeet, Whisper (Local), Qwen3 ASR and Local LLM models are unloaded when idle, after a configurable delay (10 minutes by default).
• Parakeet, Whisper (Local), Qwen3 ASR and Cohere Transcribe check the free disk space before downloading a model.

Data, updates and support

• Sync & backup can export your TypeWhisper data into one ZIP or delete all of it. The export leaves out API keys, licenses, sign-ins, models and installed plugins.
• With a commercial license, History syncs with TypeWhisper for macOS through a shared cloud folder.
• Setup can import words, corrections and snippets from Wispr Flow and custom words from Handy.
• TypeWhisper checks for updates a minute after startup and then daily, and offers Download and restart, Later or Skip this version.
• Release builds keep a diagnostic log without dictated content, and a JSON support report can be exported.

Upgrading from 1.0

TypeWhisper 1.0.9 offers 1.1.0 through its update check. On the first start without a 1.1 profile, TypeWhisper copies dictionary entries, snippets, workflows and history text, converts supported recording and text preferences, keeps the license, transfers plugin settings and encrypted API keys, installs compatible plugins and copies available model files. The 1.0 profile stays unchanged; returning to 1.0 uses it, and changes made in 1.1 are not copied back.

• Account sign-ins, archived recordings and audio, recovery audio, automation and sync settings, and settings without a 1.1 equivalent stay in the 1.0 profile.
• Missing providers are reported; TypeWhisper does not choose a replacement.
• If the local HTTP API required a token, 1.1 issues a new one. The bundled command-line client finds it on its own; scripts that used the 1.0 token need to read it again. The Raycast extension works only with Require API token turned off.
• Installing migrated plugins needs internet access.

Release candidate users

1.1.0 contains the same changes as 1.1.0-rc.2. Installations on the Release Candidate channel stay on rc.2 until the next release candidate; choose Stable as the update channel in Account & about to update to 1.1.0 and receive later stable updates.

Installation and requirements

• The installer and portable ZIP are available for x64. Setup installs the .NET 10 runtime when it is missing; the portable ZIP requires it separately.
• There is no ARM64 build anymore. ARM64 installations of 1.0 from GitHub stay on 1.0.9. Windows 11 on Arm can run the x64 version through emulation, which is untested.
• TypeWhisper requires Windows 10 build 19041 or later.
• As with 1.0.9, the GitHub installers are not code-signed, so Windows SmartScreen may ask for confirmation. The Microsoft Store version is updated separately.

Keyboard and mouse stay responsive

Fixes a system-wide input freeze where typing stopped appearing and clicks barely worked until TypeWhisper was quit (#1513).

TypeWhisper watches the keyboard for its hotkeys before other apps receive the keys. Until now every keystroke waited for TypeWhisper's main thread, so when that thread stalled, input across the whole Mac stalled with it. macOS then switched the hotkey listener off, and TypeWhisper switched it straight back on into the same stall.

• Hotkey matching now runs on its own thread. If TypeWhisper does not pick up a keystroke within 100 ms, the keystroke goes straight to the app you are typing in. A hotkey check that has already started gets up to another half second.
• While TypeWhisper is unresponsive, keys reach your apps right away. Hotkeys pressed in that time take effect once TypeWhisper responds again.
• If macOS switches the listener off repeatedly for being too slow, TypeWhisper waits before turning it back on instead of re-enabling it right away.
• Diagnostic logs now record how long TypeWhisper stopped responding.

TypeWhisper 1.7.0 is the stable 1.7 macOS release. History and Inbox now sync through iCloud, shortcuts are set on a visual keyboard, and the last dictation can be undone.

Highlights

• History and Inbox sync through iCloud. The redesigned History groups entries by device and can be filtered and searched.
• Shortcuts are configured on a visual Mac keyboard. Key labels follow the active macOS input source, and the editor shows conflicts and unavailable keys.
• The dictation indicator comes in Classic, Glass, and Light, with a live preview on the new Appearance settings page.
• Undo Last Dictation and Restore Raw Transcript are available in the menu bar and as global hotkeys. Both replace the inserted text directly and leave the clipboard untouched. If the target field or caret has changed, TypeWhisper declines and says so.
• Dictation Recovery keeps the last three successful dictations for up to 24 hours, so an incomplete provider response can be retried even when it looked successful. The Immediately retention setting turns this off.
• Files can be transcribed from Finder, and the optional Web Link plugin transcribes web media.
• Local plugins can import compatible speech models from a Hugging Face repository or a local folder. The new Canary ASR plugin is one of them.

Dictation and shortcuts

• New ways to finish a dictation: spoken submit, submit with Enter for a single dictation, Inline Commands, and an action that pastes the last transcription.
• Cancellation can be set to double Escape, single Escape, immediate, or disabled.
• Escape used to cancel a dictation no longer reaches the target app. After cancellation it works normally again.
• Fixed layout crashes in the notch indicator, the command palette, Settings, and the recent-transcriptions menu. The Overlay and Minimal indicators now use the notch indicator's safeguard.
• Fixed lost Stop hotkeys. Modifier-only shortcuts recover after event-monitor interruptions without swallowing the next Stop press.
• Automatic Fallback has an opt-in hedge: if the primary engine has not answered within a configurable threshold, the same audio also goes to the recovery engine and the first result wins. A dispatched race costs one extra API call.
• The final-transcription phase has a deadline, and dictation starts when the recovery engine is the only ready engine.
• When dictation cannot start, TypeWhisper explains why and links to the fix.
• Dictation stays pinned to the field where the recording started. TypeWhisper restores the original app and field for the final insertion and falls back to Recent Transcriptions when it cannot.
• When a failed dictation's recording was saved, the indicator says so and offers Open Recovery.
• Short, quiet dictations are transcribed in aggressive mode.
• Media is paused before Bluetooth capture starts, and the Bluetooth input is released before media resumes.
• The opt-in input preparation now covers all Bluetooth microphones and reuses the running stream between dictations, which shortens recording startup.
• Fixed external headset microphones being treated as the unavailable internal microphone in clamshell mode.
• Fixed dictation insertion in Thunderbird.
• The setup wizard distinguishes selected, installed, loaded, and tested models. Setup completes only when the current engine can transcribe and both permissions are granted. Incomplete setup offers Finish Later.

Text handling and workflows

• Number formatting offers Always, 10 and above (the default), and 100 and above under Settings → Dictation → Output Formatting. Smaller spoken whole numbers and ordinals keep their wording; decimals and recognized digit sequences still become digits. The threshold also applies when a workflow enables number normalization, and settings backups preserve it.
• Dictionary corrections apply before workflow LLM processing, and workflows share vocabulary. Long dictations can be processed in segments.
• Vocabulary can be imported from Wispr Flow, Handy, and compatible CSV files. Unsupported formats are reported as such.
• Snippet triggers require word boundaries.
• German clock times are written with a colon, and English month and day phrases are normalized. Fixed the final-period cleanup for standalone dictated values.
• Auto-submit is skipped when raw text is inserted after an LLM failure.
• When LLM post-processing fails, the raw transcription is inserted with a short notice instead of being discarded.
• An empty LLM result that two or more providers return independently is treated as intentional, not as a failure.
• Correction learning works in Electron and Chromium apps.

History, data, and administration

• Windows dictation and Recorder entries appear as their own History sources.
• Recorder rows have a copy-transcript action, and recent transcriptions sit in a workflow palette submenu.
• All app data can be exported or deleted in Advanced settings.
• typewhisper export and typewhisper import move a settings backup from the command line, with matching endpoints in the local API.
• History loads in pages with endless scrolling instead of reading the whole archive at launch.
• Fixed a finalization race in the Recorder preview and incomplete WhisperKit transcripts of Recorder sessions.
• Mac users see a pointer to the iPhone and iPad app once. It can be opened again from Settings.
• Administrators can provision ManagedLicenseKey through a macOS configuration profile or user-context script. TypeWhisper activates it automatically, reuses the activation across launches, and shows organization-managed licensing controls. See the MDM deployment guide.
• Final-transcription diagnostics include audio duration, word counts, and segment timing. They do not log audio, prompt contents, or transcript text.
• Settings use the native macOS sidebar.

Models and integrations

• Groq no longer sends dictionary terms by default. In tests with German dictations, the term list made Whisper Large V3 skip most of some dictations, while the same audio without terms came back complete. Settings you changed yourself are kept, and dictionary corrections after transcription still apply. You can turn terms back on under Groq → Send dictionary terms. A prompt sent through the HTTP API still reaches Groq when terms are off.
• On HTTP 429, OpenAI, Cohere, and plugins built on the OpenAI-compatible helpers show the provider's own message when the response contains one. Otherwise they show a generic message that names both quota and rate limits. The dictation indicator shows long error messages in full.
• The new Vercel AI Gateway plugin covers transcription and LLM processing with one API key.
• OpenAI-Compatible profiles discover Ollama models natively.
• The Gemma 4 plugin is now called Local LLM and has a new plugin ID. It gains Qwen3.5 and LFM2.5 models.
• Confucius4-R2T2 is available as a community transcription plugin.
• New plugins for Meta (Muse Voice Transcribe and Muse Spark) and Microsoft MAI Transcribe.
• Gemini adds the Gemini 3.5 Transcribe models. ElevenLabs gets settings for clean transcripts, audio-event tagging, speaker count, and dictionary keyterms. OpenAI offers only the reasoning effort levels that the selected model supports. The authenticated CLI plugin adds the free OpenCode Zen models.
• Local models of engines that are not selected are no longer restored at launch.
• Plugin HTTP requests retry transient upstream failures such as HTTP 503.
• The Parakeet plugin adds the Parakeet Ultra model. Soniox gets a free-text Transcription Context setting that is sent along with the dictionary terms.
• Parakeet no longer overcorrects with unrelated dictionary terms.
• Discover can be filtered by Local or Cloud, and TypeWhisper checks free disk space before downloading a local model.
• Plugins can provide their own pages and use authenticated command-line tools as providers.
• Parakeet restores after an automatic model unload. Deepgram Nova-3 languages no longer fall back to auto detection. An edited OpenAI API key replaces the stored one. Soniox realtime dictation stays fast without a visible preview and keeps its live transcript continuous. Fixed Deepgram live dictation streaming, a hang when uninstalling plugins, and the first Cohere model download. OpenRouter retries long audio as chunks.
• Fixed HTTP API target-language translation and session ownership.

Performance

• TypeWhisper does less work during startup, History rendering and sync, statistics updates, sound playback, and between stop and text insertion.
• The audio input callback is realtime-safe.

Compatibility and distribution

• Requires macOS 14.0 or later and supports Apple silicon and Intel Macs.
• Every release includes the iCloud bridge.
• New plugin releases from the 1.7 source line require TypeWhisper 1.7.0 or later. Existing compatible plugin releases remain available to 1.6 hosts. This app release does not republish plugins.
• Stable 1.7.0 builds use the default Sparkle channel; release candidates and daily builds remain on their preview channels.
• Simplified Chinese wording was revised.
• Built with Xcode 26.6 and Swift 6.3, with updated Swift dependencies.

Release validation is tracked in the TypeWhisper 1.7.0 checklist.

Managed license activation

Administrators can now provide a commercial license through the ManagedLicenseKey preference in the com.typewhisper.mac domain. TypeWhisper activates it on launch and saves the activation in the signed-in user's Keychain, without requiring users to enter the key or administrators to script Polar API calls.

• Reuses existing activations across launches and supports administrator key changes.
• Shows a managed license status and hides conflicting manual license controls.
• Preserves an existing license if a replacement cannot be activated.
• Suppresses license selection and purchase prompts while a managed key is configured.
• Reports Keychain storage failures instead of marking an unsaved activation as successful.

See the deployment guide and example configuration profile. Configuration profiles are not secret stores. The app does not display the key, but this does not prevent a sufficiently privileged user from retrieving it.

Bundled integration fixes

• Fix OpenRouter audio uploads for providers that require WAV instead of compressed M4A.
• Restore ElevenLabs compatibility with the TypeWhisper 1.6 host.

Broader MDM configuration for providers/models, cloud usage, updates, launch at login, and onboarding is tracked separately in #1313.

Release validation is tracked in #1315.

TypeWhisper 1.0.9 makes the latest workflow-provider plugins available to the host and fixes several issues affecting dictation, long-running workflows, and plugin management.

Highlights

• Added a configurable Filler Words plugin for removing filler words from transcribed text while preserving surrounding whitespace and punctuation.
• Added configurable US, UK, Canadian, and Australian English spelling for automatic language detection.
• Added dynamic Gemini model discovery with model-specific reasoning options and a more responsive default model.
• Made the workflow-provider plugins that require TypeWhisper 1.0.9 available in Discover, including Groq 1.0.6.

Reliability improvements

• Kept dictation hotkeys responsive and fixed low-level hook initialization on their dedicated message-loop thread.
• Unloaded local transcription models when switching providers, so inactive models no longer retain their resources.
• Brought the existing TypeWhisper window to the foreground when the app is launched a second time.
• Handled unavailable clipboard bitmap and file-drop formats without blocking dictation.
• Prevented long online transcription previews from stalling while preserving rolling preview cadence, punctuation, and unspaced CJK text.
• Increased output budgets for long workflow prompts and rejected explicitly truncated provider responses instead of returning incomplete text.
• Filtered high-confidence terminal “Thank you” hallucinations from Groq transcripts while preserving genuine dictated content.

Plugins and packaging

• Fixed the manual plugin folder shown by Microsoft Store builds so the path is visible to Explorer and the Open folder button reaches the real MSIX data directory.
• Made plugin registry publishing resilient to overlapping releases and kept Filler Words versioning independent from app releases.
• Added automated Windows UI screenshot capture and expanded release metadata validation for workflow-provider plugins.

Validation notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -daily suffixes.
• Validate both published portable ZIPs and record the Authenticode status of the published installers.
• Publish Microsoft Store package 1.0.9.0 and WinGet package 1.0.9 separately from the GitHub release.

TypeWhisper 1.6.0 is the stable 1.6 macOS release. It brings a redesigned settings experience, privacy-friendly statistics and portable backups, faster and more resilient recording, calendar-aware meeting automation, expanded workflow integrations, and a broad reliability pass across dictation, Recorder, local models, and cloud providers.

Highlights

• Unified the visual language across Settings, including dashboards, forms, collections, workspaces, and advanced tool pages.
• Added a Statistics view for privacy-friendly aggregate usage insights.
• Added Backup & Restore for portable workflows, dictionary entries, snippets, profiles, prompt actions, hotkeys, installed community plugins, transcription history text, update-channel selection, and supported preferences.
• Improved recording startup performance in controlled hardware tests: median request-to-first-buffer time fell by 22.6% on the built-in MacBook microphone and 13.1% on a tested USB microphone, while opt-in AirPods recording readiness improved by about 45%.
• Added optional calendar-aware meeting automation with notifications, countdown controls, meeting activity checks, automatic Recorder sessions, sortable filenames, versioned transcript metadata, and Obsidian-ready Markdown transcript sidecars for the associated event.
• Added an MCP Client action add-on with stdio and Streamable HTTP support, expanded Obsidian with live sync and Markdown templates, and added bulk plugin updates with automatic restart.
• Added Simplified Chinese localization across the app and bundled integrations; two AirPods keep-alive descriptions remain in English.

Dictation and Recorder

• Added an ordered microphone priority list and improved recovery for AirPods, Bluetooth system-default input, clamshell changes, virtual inputs, and FaceTime built-in microphone capture.
• Moved recording startup work off the main actor and prewarmed eligible inputs to reduce the delay before audio capture begins.
• Improved Whisper Mode amplification with smoother gain changes, less amplification of low-level background noise, stable handling of brief pauses, and protection against clipping; standard recording mode remains unchanged.
• Added optional live transcript updates to active text fields while preserving the final insertion path.
• Fixed automatic meeting transcripts to use the finalized recording as their source of truth, with a captured-audio fallback when the saved file cannot be read.
• Improved long-meeting transcription completeness by aligning final Recorder transcription and retries with File Transcription's source-progress-aware path.
• Improved contextual insertion for Firefox, CJK text, trailing spaces, secure-input situations, auto-submit targets, and target-app correction learning.
• Improved indicator placement, dismissal, fullscreen visibility, feedback interaction, countdown progress, and optional screen-capture visibility.
• Added recovery-audio retention controls and improved direct Recorder retranscription when local models have unloaded.

Dictionary, Workflows, and Automation

• Added dictionary search, safe reset actions, grouped correction variants, clearer learned-correction visibility, and a guided microphone correction trainer.
• Applied dictionary corrections to file transcription and connected per-term CTC tuning to Parakeet.
• Added ordered global LLM provider and model fallbacks while keeping explicit workflow providers strict.
• Added explicit workflow dictation through the local API and a live-preview engine override so preview can stay local while final transcription uses the configured dictation engine.
• Regrouped the menu bar for faster access and now shows the active transcription provider alongside the selected model.

Models and Integrations

• Added Soniox live WebSocket transcription and the local Cohere Transcribe integration.
• Added context-aware OpenAI transcription and dynamic ChatGPT model discovery.
• Expanded OpenAI-Compatible profiles with optional API versions, deployment-scoped endpoints, realtime transcription, and Responses API support.
• Updated the separately distributed OpenRouter plugin to preserve segment timestamps for SRT and VTT exports on supported speech-to-text routes; Deepgram through OpenRouter remains text-only.
• Improved MLX model storage, lazy restore, stalled-download recovery, and lifecycle handling across Qwen3, Gemma 4, Granite, Voxtral, Parakeet, and WhisperKit.
• Fixed Qwen3 long-audio cutoff and loop-fallback behavior, and improved WhisperKit restore after unloads and for workflow-specific model overrides.
• Improved Soniox cleanup and quota reporting, Speechmatics US batch routing and language-pack handling, restricted ElevenLabs key validation, AssemblyAI file-transcription routing, Mistral language handling, and Deepgram custom-URL validation.

Platform and Distribution Notes

• Fixed calendar permission requests and the calendar entitlement used by distributed builds.
• Fixed the Settings sidebar scroll pocket on macOS 27 while preserving the existing layout on macOS 14 through 26.
• Stable 1.6.0 builds use the default Sparkle channel and require macOS 14.0 or later; release candidates and daily builds remain on their dedicated preview channels.
• The final stable tag publishes DMG and ZIP assets, updates the stable Sparkle appcast entry, triggers the website release update, and updates the Homebrew cask.
• Automatic private iCloud sync remains disabled while the production container and Developer ID provisioning profile are unavailable. Cloud Folder Sync through a user-selected iCloud Drive, Dropbox, OneDrive, Syncthing, or custom folder remains available.

Release validation is tracked in the TypeWhisper 1.6.0 checklist.

TypeWhisper 1.0.8 is a stable Windows feature and reliability release focused on more flexible workflows, safer authenticated provider integrations, and stronger state preservation across recording, updates, and clipboard use.

Highlights

• Added a first-party integration for authenticated provider CLI sessions used by LLM workflow processing, with executable discovery, authentication status, structured input and output, timeouts, cancellation, and process cleanup.
• Added configurable App or website and App and website matching for automatic workflow triggers while preserving the behavior of existing combined workflows.
• Added \s support for inserting literal spaces in dictionary replacements, with updated guidance in every supported app language.

Reliability Improvements

• Preserved explicitly disabled term packs, vocabulary boosting preferences, and pending industry presets across updates, catalog refreshes, licensing, and user-data migration.
• Fixed recorder overlays so finalizing is no longer shown as an active recording, with more reliable placement across monitors and DPI scales.
• Improved bitmap clipboard capture and restoration for delayed, unavailable, and OLE bitmap representations during temporary paste operations.
• Kept missing or unavailable provider selections visible and actionable instead of silently routing workflow processing to another provider.

Security and Compatibility

• Restricted provider processes to fixed arguments, a controlled environment and working directory, bounded output, disabled tools and project context, and process-tree termination on cancellation or timeout.
• Kept unsupported provider paths unavailable until equivalent per-invocation safety controls can be verified.
• Added stable UI Automation identifiers and clearer recorder overlay status descriptions for accessibility and runtime validation.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -daily suffixes.
• Validate both published portable ZIPs and record the Authenticode status of the published installers.
• Publish Microsoft Store package 1.0.8.0 and WinGet package 1.0.8 separately from the GitHub release.

TypeWhisper 1.0.7 is a stable Windows feature and reliability release focused on better transcription output, faster recovery, safer local models, and more capable integrations.

Highlights

• Added Smart Formatting, regular-expression dictionary rules, short-utterance punctuation, spoken-number normalization, and regional German output for Germany, Austria, Switzerland, and neutral written German.
• Added interactive usage statistics with daily activity, language and app breakdowns, keyboard-accessible details, and retention-independent totals.
• Improved immediate dictation startup, automatic recovery after interrupted sessions, and clipboard restoration when text insertion fails.
• Added unified local-model download requirements, retry and removal controls, optional Hugging Face authentication, and revision-aware license acceptance.

Integrations and Models

• Added local Cohere Transcribe support and hardened the packaged Sherpa ONNX, cuDNN, Whisper.cpp, Granite Speech, and Supertonic TTS runtimes.
• Added recorder-specific transcription engine and model overrides for long recordings without changing the main dictation model.
• Added configurable request timeouts per OpenAI Compatible profile and safer Unicode, reasoning, and thinking-content handling.
• Redesigned Integrations around Installed, Discover, and Manual sources with trusted registry artifacts, community feeds, compatibility diagnostics, update-state details, and repair actions.
• Added provider-aware dictionary budgets and stronger workflow input and output boundaries for plugin integrations.

Windows Experience

• Added Simplified Chinese localization and expanded localized model, integration, and recorder controls.
• Added an About section with project, support, documentation, and license links.
• Displayed History timestamps in local time while preserving UTC storage.
• Reworked Script Runner editing and made mouse-based dictation hotkeys safer.
• Improved onboarding model update state and local-model acceleration details.

Reliability Improvements

• Serialized recorder start and stop transitions so concurrent API requests cannot overwrite active sessions or leak model scopes.
• Made local-model progress, cancellation, removal, selection persistence, and service disposal safe across concurrent operations.
• Hardened plugin discovery, compatibility checks, package verification, rollback, path validation, and broken-install reporting.
• Preserved existing text and selection state across workflow processing and insertion recovery paths.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -rc or -daily suffixes.
• Validate both published portable ZIPs and record the Authenticode status of the published installers.
• Publish Microsoft Store package 1.0.7.0 and WinGet package 1.0.7 separately from the GitHub release.

TypeWhisper 1.0.6 is a stable Windows performance and reliability release focused on faster startup, responsive settings, dependable microphone capture, and safer local transcription.

Highlights

• Rendered the tray, recording overlay, and settings shell before slower plugin, model, and audio-device initialization work.
• Replaced periodic audio-device polling with event-driven notifications and cached enumeration.
• Restored low-latency WASAPI recording with automatic WaveIn fallback while continuing to release capture resources between dictations.
• Added isolated CUDA runtime safety checks with automatic CPU fallback for local transcription.
• Increased the Individual license tier from two to three devices.

Integration Updates

• OpenAI plugin 1.1.1 adds gpt-transcribe and gpt-live-transcribe, dynamic account-specific model catalogs, ordered language hints, dictionary terms, and safer cached fallbacks.
• Soniox plugin 1.1.0 uses compressed M4A uploads when available and retries with the original WAV when compression is unavailable or rejected.
• Both integration updates are available through the TypeWhisper plugin marketplace and remain optional.

Reliability Improvements

• Deferred local model loading until first use to reduce startup work.
• Rolled back cancelled plugin initialization cleanly.
• Hardened microphone device changes, capture startup, and fallback paths.
• Preserved valid OpenAI model selections when remote catalogs are sparse or temporarily unavailable.
• Matched the three-device Individual entitlement exactly without misclassifying larger device counts.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -rc or -daily suffixes.
• Validate both published portable ZIPs and record the Authenticode status of the published installers.
• Publish Microsoft Store package 1.0.6.0 and WinGet package 1.0.6 separately from the GitHub release.

TypeWhisper 1.0.5 is a stable Windows maintenance release focused on reliable dictation sessions, safer portable upgrades, clearer plugin configuration, and better local integrations.

Highlights

• Released microphone capture between dictations so consecutive recordings use a fresh capture session.
• Added safe cross-volume migration for legacy portable user data.
• Added workflow-specific dictation through the local API while preserving existing bodyless start requests.
• Added configurable ElevenLabs transcription modes, audio-event handling, speaker options, and TypeWhisper dictionary support.

Improvements

• Moved plugin settings into accessible, resizable modal windows that open after installation and from installed plugin cards.
• Added staged Save and Cancel behavior to the Script Runner editor.
• Exposed active workflow details and workflow dictation capability through API status responses.
• Added x64 and ARM64 portable archive validation to package dry runs and release builds.

Fixes

• Fixed side-specific bare-modifier main dictation hotkeys such as Right Alt and Right Ctrl.
• Fixed microphone resources remaining open between recording sessions.
• Fixed portable startup when legacy user data resolves to another volume.
• Fixed Script Runner edits collapsing the selected script after each input.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -rc or -daily suffixes.
• Validate both published portable ZIPs and record the Authenticode status of the published installers.

New Features

• select data-residency region (US/EU/JP) with auto-detect

Bug Fixes

• disable API key box during region auto-detect probe
• bound region auto-detect probes and lock region UI during test

Other Changes

• Preserve chat helper binary compatibility
• Hide Groq Qwen reasoning output
• Update GraniteSpeech PyTorch dependency
• Enable autostart for new installations
• Harden file transcription test isolation
• Add file transcription regression coverage
• Bump Live Transcript plugin to 1.0.3
• Prevent duplicate transcript entrance animation
• Address correction trainer review feedback
• Stabilize live transcript and recording overlays
• Add guided correction trainer
• Address settings automation review feedback
• Fix dictionary pack state notifications
• Refine recording overlays and live transcript
• Add settings UI automation coverage
• Simplify dictionary reset actions
• Persist correction provenance and add safe dictionary resets
• Retry restored code quality service
• Retry transient code quality check
• Improve correction group accessibility
• Refine correction group layout
• Group dictionary corrections by replacement
• Handle invalid diagnostics roots consistently
• Recover missing correction learning target window
• Refresh localization resources on build
• Improve correction learning visibility
• Preserve external HTTP timeout behavior
• Handle stalled plugin downloads
• Add privacy-friendly purchase attribution (#287)
• Strengthen hybrid modifier regression test
• Fix hybrid modifier hold behavior
• Require double Escape to cancel dictation
• Address review feedback
• Add ordered language hints
• Address review feedback
• Preserve hotkey recorder clear action
• Preserve workflow provider during refresh
• Protect user data across release channels
• Refine Windows hotkey handling
• Allow workflows without post-processing
• Serve plugins from registry origin
• Publish plugin package hashes
• Unsubscribe wizard close handler
• Clarify Parakeet plugin updates
• Fix wizard lifetime when settings close
• Address Store packaging review feedback
• Add Microsoft Store packaging workflow
• Address microphone priority review feedback
• Stabilize microphone device priority selection
• Keep uninstall audio recovery non-interactive
• Address recording protection review feedback
• Protect user recordings during uninstall
• Address review: follow default-endpoint changes and honor explicit preview device
• Follow the system default capture device in automatic microphone mode
• await background cleanup in the two auto-language tests
• run best-effort cleanup off the transcription critical path
• use current stt-async-v5 model
• poll async status with adaptive backoff

TypeWhisper 1.0.1 is a stable Windows maintenance release. It focuses on local speech-engine reliability, cleaner plugin startup behavior, and development dashboard tooling for support and QA workflows.

Highlights

• Promoted the latest 1.0.1 maintenance fixes to the stable Velopack channel.
• Improved Sherpa-ONNX native runtime loading for local transcription setups.
• Kept optional plugins opt-in so a fresh app install starts with only bundled capabilities active.

Improvements

• Added development dashboard seed controls to make support, QA, and demo data scenarios easier to reproduce.
• Refined development seed busy-state handling and test cleanup.
• Stopped installing marketplace plugins by default during first-run setup.

Fixes

• Fixed Sherpa-ONNX native runtime probing so required runtime DLLs are resolved from the plugin package.
• Fixed Sherpa-ONNX CUDA runtime aliasing for systems that use accelerated local transcription.
• Addressed review feedback around Sherpa-ONNX runtime loading behavior.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -rc or -daily suffixes.
• Check installer Authenticode status explicitly before announcing release quality.

Other Changes

• Fix WhisperKit restore wait for long model loads

TypeWhisper 1.0.0 is the stable Windows 1.0 release. It brings the 1.0 release candidate work to the stable update channel with update reliability, plugin lifecycle fixes, target-app learning improvements, and the Windows 1.0 screenshot set.

Highlights

• Promoted the Windows 1.0 release line from release candidate to the stable Velopack channel.
• Kept the stable release focused on reliability, update behavior, plugin lifecycle, and release polish.
• Added the Windows 1.0 screenshot set for the current app surfaces.

Improvements

• Improved update-channel behavior for explicit stable, release candidate, and daily channel selections.
• Improved bundled plugin uninstall handling so app-bundled plugin directories are removed consistently.
• Improved target-app autocorrection learning for Electron-based apps.
• Included plugin script files in release packages and refreshed bundled plugin compatibility for the 1.0 line.

Fixes

• Fixed bundled plugins returning after restart after uninstall.
• Fixed daily release version-base resolution after RC/stable tags.
• Fixed Groq recorder transcription on systems without Media Foundation AAC by falling back to WAV upload.
• Resolved Granite Speech package audit issues for release packaging.

Validation Notes

• Verify stable Velopack channels win-x64 and win-arm64.
• Confirm x64 and ARM64 installers, portable ZIPs, full nupkg packages, RELEASES files, and JSON manifests are present without -rc or -daily suffixes.
• Check installer Authenticode status explicitly before announcing release quality.

TypeWhisper 1.5.0 is the stable 1.5 macOS release. It brings app-aware dictation insertion, virtual audio input support, broader number normalization, expanded bundled speech providers, local-model memory controls, dictionary-learning improvements, and a focused reliability pass across hotkeys, workflows, plugins, model handling, and release metadata.

Highlights

• Added app-aware smart insertion so dictated text better respects surrounding text, sentence position, trailing spaces, terminal paste behavior, and target-app context.
• Added virtual audio input device support and hardened media pause/resume behavior, recording start feedback, mouse-button shortcuts, and fullscreen indicator handling.
• Expanded number normalization across generic transcription output, multilingual number words, Dutch number words, English ordinals, English digit sequences, and French decimal phrases.
• Added and refreshed bundled speech and AI providers, including Gemini speech transcription, Cartesia speech transcription, Sber SaluteSpeech, OpenRouter speech-to-text, Reson8, Mistral AI, OpenAI-compatible profiles, Soniox regions and TTS, and updated cloud ASR providers.
• Added Japanese localization and dictation support, ordered language hints, recent transcriptions in the workflow palette, source progress for file transcription, recorder-specific engine overrides, and a short-dictation workflow AI-skip setting.
• Added local MLX memory-footprint controls, idle local-model auto-unload behavior, and recovery for stalled Gemma 4 model downloads.
• Added pro transcription fallback, auto-learned dictionary corrections, per-term dictionary CTC tuning plumbing, and tighter target-app correction learning.

Reliability and Fixes

• Fixed delayed Hybrid hotkey behavior, restored non-Control hybrid modifier taps, improved global push-to-talk capture, aligned Pages workflow hotkeys with palette behavior, added menu-based dictation pause, and required confirmation before Esc cancels dictation.
• Fixed fullscreen indicator menu-bar strip issues and the top overlay menu-bar strip.
• Fixed Groq and cloud ASR compressed M4A upload handling, including a WAV fallback for incomplete M4A finalization.
• Fixed Soniox async file transcription routing, region persistence, and realtime model selection.
• Fixed OpenAI Compatible GPT-5 token parameter handling, host SDK compatibility, and configurable request-timeout behavior.
• Fixed plugin host compatibility metadata, raised cloud ASR plugin host floors to TypeWhisper 1.5 where required, and prevented plugin releases from taking over the repository-level GitHub Latest marker.
• Fixed TaskForge preserve-clipboard insertion fallback, dictionary scroll crashes, event tap teardown, FaceTime built-in microphone capture, plugin uninstall keychain cleanup, and development build cleanup scope.
• Fixed Parakeet v2/v3 model setup by repairing missing vocabulary downloads before model loading.
• Improved Japanese dictation post-processing and settings localization, Latin word replacement boundaries, terminal clipboard restore verification, and recorder final transcription failure surfacing.

Developer and Distribution Notes

• Stable 1.5.0 builds use the default Sparkle channel and require macOS 14.0 or later.
• Release candidates and daily builds remain on their dedicated Sparkle channels and do not update Homebrew.
• The final stable tag publishes DMG and ZIP assets, updates the stable Sparkle appcast entry, triggers the website release dispatch, and updates the Homebrew cask.
• The 1.5 plugin line uses the community-capable registry feed and keeps host compatibility metadata aligned with the bundled plugin SDK.

Other Changes

• Add recorder system audio parity (#233)
• Fix Groq recorder long transcription
• Address plugin update review feedback
• Fix plugin update state handling
• Add Chinese spoken language option
• Fix CJK dictionary corrections
• Fix Python package audit for unresolved Torch advisory
• Fix overlay recovery after display changes
• Release Live Transcript plugin 1.0.2
• Handle stale provider profile selections
• Add OpenAI-compatible provider profiles
• Add number normalization parity
• Address static analysis review findings
• Address review feedback for translation storage flow
• Address review feedback for storage paths
• Address review feedback for asset paths
• Add quick translation mode and model storage controls
• Add file transcription queue clearing
• Document public production APIs
• Add multiple workflow hotkeys
• Release Soniox plugin 1.0.3
• Rework license management flows
• Group Soniox transcript tokens for subtitles
• Add Premium Cloud Folder Sync
• Add AMD acceleration for whisper.cpp
• Require restart after CUDA runtime installation

Other Changes

• Fix CUDA runtime delivery for local plugins
• Fix Soniox API key review issues
• Address Soniox review feedback
• Fix Soniox async transcription
• Add smart trailing space after dictation
• Add Reson8 to plugin release workflow (#178)
• Add Reson8 transcription plugin (#177)
• Marshal workflow fallback feedback to UI thread
• Fix workflow post-processing fallbacks
• Guard dictation audio history filenames
• Harden dictation persistence follow-ups
• Address dictation output persistence review
• Preserve preview fallback for long dictation
• Address writing focus review feedback
• Add acceleration restart prompt (#172)
• Fix writing focus localization
• Harden localization and live transcript tests
• Fix live transcript appearance settings
• Harden Live Transcript terminal event matching
• Fix Live Transcript no-speech cleanup
• Fix Windows-key hotkey focus handling
• Add SignPath application documentation
• Avoid blocking streaming audio callbacks
• Serialize streaming audio sends
• Fix online live transcription behavior
• Stabilize Granite Speech dependency audit
• Handle dispatcher shutdown safely
• Harden dispatcher fallback during tests
• Stabilize plugin system test run
• Address acceleration reload review feedback
• Address remaining review nits
• Tighten review follow-up handling
• Fix CUDA acceleration reload
• Stabilize review check follow-ups
• Address HTTP API review findings
• Stabilize settings dispatcher events
• Stabilize HTTP API parity checks
• Align Windows HTTP API with Mac contract
• Harden workflow prompt tests
• Frame workflow LLM input as data
• Remove issue Discord relay workflow (#153)
• Add Smallest AI plugin release mapping
• Add Smallest AI Pulse transcription plugin
• Fix onboarding completion for unready transcription engines (#148)
• Fix Groq API key setup in onboarding (#147)
• Add OpenAI model parity and default provider clarity

TypeWhisper 1.4.0 is the stable release for the new Integrations and community-plugin foundation. It brings the redesigned plugin hub, the 1.4 community registry feed, downloaded model management, new local automation APIs, and a broad reliability pass across workflows, live transcript, indicators, audio capture, and first-party plugins.

Highlights

• Added the plugins-community-v1.json registry feed for 1.4 builds while keeping the 1.3.x line on the existing official marketplace feed.
• Redesigned Integrations with Installed, Discover, and Manual tabs, grouped plugin lists, source and hosting metadata, and capability filtering for local, cloud, transcription, LLM, action, and memory plugins.
• Added support for community plugin registry publishing and mirrored first-party plugin releases into the community registry feed.
• Added downloaded model management for local model plugins, including host-version gating for the Qwen3, Granite, Voxtral, Gemma 4, WhisperKit, and Supertonic release line.
• Added the Smallest AI Pulse transcription plugin, the xAI Grok plugin, the experimental Supertonic TTS plugin, and refreshed OpenAI voice and Qwen3 ASR integrations for 1.4.
• Added local Recorder and Dictionary HTTP API endpoints so trusted local clients can inspect recorder state, start transcriptions, and manage dictionary terms through the app API.
• Added watch-folder file job automation, per-dictation STT engine/model overrides, workflow drag reordering, and multiple global hotkeys per action.
• Added Parakeet live transcription support and the WhisperKit Distil Large v3 Turbo model.
• Added licensed industry term packs and localized built-in term pack metadata.

Reliability and Fixes

• Tightened workflow input boundaries so dictated text and Apple Intelligence scaffold output are treated as source content instead of executable instructions.
• Improved live transcript stability, sentence accumulation, correction handling, preview latency, and maximized-window behavior.
• Kept recording indicators visible for active recorder sessions while hiding supported indicator windows from screen captures.
• Improved fullscreen, fixed-display, foreign-window, and maximized-window indicator handling.
• Reduced hotkey recording startup latency and fixed exact modifier-combination matching.
• Recovered failed dictation audio and added audio diagnostics export for troubleshooting capture failures.
• Improved selected USB microphone capture, multichannel input handling, and system-audio recorder sample handling.
• Deferred the recording start cue until input is ready and improved Parakeet onboarding/setup status.
• Made plugin settings windows scrollable and restored workflow action targets.
• Improved onboarding layout, transcription helper compatibility, memory extraction for normal dictations, and plugin registry release metadata.
• Hardened webhook sensitive header storage and local API token enforcement.

Developer and Distribution Notes

• The stable Sparkle appcast entry for 1.4.0 uses the default channel and requires macOS 14.0 or later.
• Homebrew updates are intentionally stable-only and are triggered by the final v1.4.0 tag.
• Release candidates and daily builds remain on their dedicated Sparkle channels and do not update Homebrew.
• Plugin SDK workflow snapshots expose read-only workflow metadata without handing plugins internal app model objects.

TypeWhisper 0.8.1 is a focused Windows release that expands cloud provider coverage, improves plugin marketplace reliability, and fixes several dictation and local transcription issues found after 0.8.0.

New Features

• Added OpenRouter model catalog support so compatible models can be discovered and selected more easily.
• Added OpenRouter transcription support for cloud dictation workflows.
• Added the xAI Grok provider plugin with streaming transcription, text generation, and TTS support.
• Added branded plugin logos for major bundled providers.

Improvements

• Hardened plugin registry category handling so marketplace entries stay compatible as categories expand.
• Made the Live Transcript window movable and improved its settings experience.
• Improved local GPU acceleration reload behavior for local transcription providers.

Fixes

• Fixed final decoding for cloud dictation responses.
• Fixed Live Transcript failure handling so provider errors are surfaced more reliably.
• Improved model acceleration layout coverage and related plugin system behavior.

TypeWhisper 0.8.0 is a major Windows release focused on faster local transcription, a broader plugin marketplace, richer speech and TTS provider support, and steadier everyday dictation behavior.

New Features

• Added GPU acceleration for local transcription so supported local models can run faster on capable Windows systems.
• Added multi-category plugin marketplace support, making discovery clearer as plugins expand across transcription, text generation, TTS, memory, and workflow integrations.
• Added the Supertonic TTS provider plugin and a dedicated TTS marketplace category.
• Added side-specific modifier hotkeys for workflows that need distinct left/right keyboard shortcuts.
• Added Russian localization for the Windows app.
• Added Windows indicator style variants for a more flexible recording status experience.
• Added licensed industry term packs for specialized vocabulary workflows.
• Expanded OpenAI and ChatGPT plugin capabilities.

Improvements

• Improved plugin marketplace behavior and polish, including category handling and uninstall state feedback.
• Improved whisper.cpp native runtime packaging so bundled local transcription dependencies are more reliable.
• Improved touch scrolling across ScrollViewer controls.
• Limited Supertonic TTS runtime packaging to keep bundled plugin output cleaner.

Fixes

• Fixed microphone reconnect handling.
• Reduced repeated short transcription phrases and improved dictation ellipsis cleanup.
• Fixed plaintext insertion for email clients.
• Fixed live preview font size updates.
• Fixed live transcript no-speech handling.
• Fixed dictionary usage count localization.
• Fixed marketplace plugin uninstall state handling.

TypeWhisper 1.3.3 is a stable hotfix release for the 1.3 line. It fixes selected USB audio interfaces still being marked as not compatible before preview or recording can start.

Bug Fixes

• Fixed USB microphones and audio interfaces, including RME Babyface-style devices, being blocked by selection-time compatibility checks.
• Kept non-Bluetooth input selection non-blocking so TypeWhisper can reach the input-only HAL preview and recording path added in 1.3.2.
• Preserved Bluetooth route activation, stabilization, and aggregate-route validation behavior.

Testing Focus

• If a USB interface still showed (Not compatible) in TypeWhisper 1.3.2, test selecting it again in 1.3.3.
• After selection, use Test Microphone to confirm the device now reaches preview capture.

TypeWhisper 1.3.2 is a stable hotfix release for the 1.3 line. It fixes selected USB audio interfaces being marked as not compatible when the selected input and current output device use different clocks or sample rates.

Bug Fixes

• Fixed selected USB microphones and audio interfaces that could show (Not compatible) before preview or recording started.
• Routed explicit non-Bluetooth input devices through an input-only CoreAudio HAL capture path, avoiding stale AVAudioEngine full-duplex formats when input and output clocks differ.
• Kept the existing AVAudioEngine route for system default and Bluetooth inputs, preserving the established Bluetooth stabilization and recovery behavior.
• Downmixed multi-channel input buffers to mono before the existing 16 kHz transcription conversion path.

Testing Focus

• If a USB interface such as an RME Babyface, Elgato Wave XLR, or similar device showed (Not compatible) in TypeWhisper 1.3.1, test selecting it again in 1.3.2.
• If the device still fails, also test with the interface set as the macOS default input and with input/output sample rates aligned at 48 kHz.

TypeWhisper 0.7.2 is a stability and polish release for Windows. It improves dictation reliability, adds Japanese localization, makes the preview bubble easier to control, and hardens the newest workflow and paste behavior from the 0.7 line.

New Features

• Added Japanese localization for the Windows app.
• Added an auto-hide setting for the preview bubble.
• Added manual workflow controls and advanced workflow triggers on Windows.

Improvements

• Kept the preview bubble compact and scrolled so longer live transcription output stays easier to follow.
• Improved Parakeet transcription tail handling with optional tail padding, better final-result selection, and internal diagnostics for clipped endings.
• Added audio tail telemetry around recording stop so hard-to-reproduce tail clipping can be diagnosed more clearly.

Fixes

• Reduced repeated transcription phrases in final dictation output.
• Fixed paste handoff latency so inserted text is less likely to race with the target app.
• Preserved clean no-speech handling for short and quiet clips while improving final transcription padding.

Other Changes

• Fix mouse-mapped hotkeys on Windows (#84)
• Fix/escape cancel stuck state (#85)
• Fix Windows SendInput struct layout for auto-paste (#83)
• Fix Escape cancel hotkey registration (#82)

TypeWhisper 1.3.1 is a stable maintenance release for the 1.3 line. It brings the workflow improvements from the 1.3.1 release candidates to stable users, adds direct workflow text hotkeys and a spoken-language dictation template, improves plugin metadata, and includes reliability fixes for Deepgram code-switching, Bluetooth audio, indicators, local LLM cleanup, and large automation requests.

New Features

• Added an Apple Translate workflow processor for local translation workflows on supported macOS configurations.
• Added a manual workflow trigger so workflows can stay out of automatic dictation matching and run from the Workflow Palette instead.
• Added a recorder toggle in the menu bar for faster access to app audio recording.
• Added packaging support for the bundled Filler Words plugin.
• Added a workflow dictation template for spoken-language transformations.
• Added direct workflow text hotkeys so workflows can run on selected or supplied text without starting a dictation.

Bug Fixes

• Exposed Deepgram multilingual code-switching support so mixed-language dictation can use the provider capability correctly.
• Reduced workflow post-processing latency so prompt and processor workflows return results faster after transcription.
• Wrapped Apple Intelligence workflow input so dictated text is treated as source content instead of executable instructions.
• Stabilized Bluetooth dictation capture and guarded AirPods output volume during audio routing changes.
• Opted indicators out of fullscreen spaces to avoid unexpected overlay behavior when apps use fullscreen mode.
• Restored indicator visibility across macOS spaces.
• Disabled automatic Live Transcript opening by default.
• Auto-unloaded local LLM post-processing resources after use to reduce background memory pressure.
• Supported large CLI transcriptions without truncating or failing oversized requests.
• Kept the language list available before transcription plugins finish loading.
• Fixed a Parakeet live preview fallback race.

Other Changes

• Classified plugin hosting explicitly so local and cloud plugins appear with clearer metadata in plugin listings.
• Documented plugin hosting metadata in the SDK manifest guidance.
• Updated Parakeet vocabulary boosting capability detection for current local-model behavior.
• Expanded automated coverage for workflow text hotkeys, spoken-language workflow templates, plugin hosting compatibility, and Deepgram multilingual code-switching.

TypeWhisper 0.7.0 is a major workflow and automation release for Windows. It brings the workflow system into the main settings experience, expands file transcription automation, adds new provider capabilities, and tightens dictation reliability for everyday use.

New Features

• Added the unified Workflows surface for prompt actions, matching rules, workflow-specific hotkeys, language and model overrides, output formatting, and action routing.
• Added batch file transcription queues with per-file status tracking, export options, and improved model selection behavior.
• Added watch folder transcription so folders can be monitored and processed automatically with configurable export output.
• Added the recent transcriptions palette for quickly finding, editing, copying, and reusing recent dictation results.
• Added the ElevenLabs transcription plugin, including streaming support and localized settings.
• Added TTS provider plugin support, including host services, playback sessions, voice metadata, and a built-in Windows SAPI provider.
• Added stable, release-candidate, and daily update channel support in the app and release packaging.

Improvements

• Expanded the local HTTP API and bundled CLI with per-request engine and model overrides, language hints, translation targets, dictionary term management, and dictation controls.
• Improved automation compatibility for external tools through richer API parsing, status responses, and CLI installation support.
• Improved file transcription handling for larger queues, subtitle export, media formats, and watch folder output.
• Improved plugin marketplace and plugin host behavior so bundled and external plugins can expose more capabilities consistently.
• Improved short and quiet speech handling with configurable padding and gain behavior.
• Refined README, issue templates, PR templates, package dry runs, plugin smoke checks, and release workflow coverage.

Fixes

• Fixed dictation tail clipping so the end of short recordings is less likely to be cut off.
• Hardened automatic paste reliability after transcription completes.
• Fixed prompt provider selection and related history/prompt provider race conditions.
• Fixed autostart toggle handling.
• Hardened watch folder exception handling so background folder processing is more resilient.

TypeWhisper 1.3 is the stable macOS release centered on the new Workflows experience, more capable automation surfaces, and a broad round of dictation, UI, and local-model reliability work across the 1.3.0-rc* cycle.

Highlights

• Unified Workflows - TypeWhisper now brings prompt actions and matching rules into one dedicated Workflows surface, with a native editor, wizard flow, and a global Always fallback trigger.
• Better automation control - The local HTTP API and CLI now support per-request STT engine and model selection, while multilingual language hints make mixed-language dictation easier to steer.
• More polished dictation feedback - Spoken feedback, adjustable transcript sizing, faster recent-transcription recovery, and grouped menu bar actions make day-to-day use smoother.
• Stronger local-model path - The 1.3 cycle hardens local plugin downloads, Hugging Face token handling, model restore behavior, and Parakeet language filtering.

Workflows & Prompt Processing

• Replaced the old Prompt Actions and Rules settings surfaces with the new unified Workflows experience.
• Added a native macOS sidebar and wizard-style create/edit flow for workflows.
• Added a global Always workflow trigger that acts as the fallback after website and app-specific matches.
• Hardened workflow prompts so dictated text is always treated as source text to transform, not instructions to execute.
• Added per-workflow temperature controls for supported LLM providers.
• Preserved focus in the original target app during local prompt processing.

Dictation, Feedback & UI Polish

• Added spoken feedback through the bundled System Voice plugin, scoped to transcription readback only.
• Added multilingual language hints with multi-select search and selected-count feedback.
• Added shortcuts for reopening recent transcriptions and a recovery palette for recent results.
• Added adjustable live transcript preview sizing for dictation overlays.
• Improved the notch and overlay indicator behavior, including active-screen re-anchoring and timer layout fixes.
• Grouped menu bar actions into clearer sections for faster navigation.
• Fixed Fn hotkey handling so both press-and-release and press-and-hold strategies work reliably.

Local Models, Plugins & Reliability

• Upgraded the local model runtime and download flows for bundled on-device plugins.
• Added shared Hugging Face token handling for the bundled local-model plugins and clearer validation where required.
• Fixed Parakeet language filtering against the current FluidAudio upstream behavior.
• Improved WhisperKit restore and unload handling.
• Split the marketplace feeds by compatibility line and hardened plugin loading so incompatible external bundles are skipped instead of breaking launch.
• Made dictionary prompts engine-aware and kept streaming dictionary terms working across AssemblyAI, Soniox, and SpeechAnalyzer.
• Added subtitle exports for Watch Folder jobs.

Power Users & Automation

• Added per-request STT engine and model selection through the local HTTP API and the typewhisper CLI.
• Kept the documented /v1/* API and CLI command surface stable for the 1.x line.
• Improved plugin compatibility validation and diagnostics for marketplace and bundled plugins.

TypeWhisper 0.6.1 is a focused stability release that tightens dictation behavior, file transcription model selection, provider state handling, and release packaging.

Improvements

• Improved provider and model state handling so the app is less likely to lose the selected configuration while switching between engines and prompts.
• Refined the bundled Groq integration with better LLM model selection and refresh behavior.

Fixes

• Fixed recorder-compatible hotkey parsing so captured shortcuts round-trip correctly through the UI and runtime hotkey pipeline.
• Fixed file transcription to load the model the user actually selected instead of falling back unexpectedly.
• Fixed the silent-stop overlay reset path and closed a streaming session leak that could leave overlay state behind after recording ended.

Release Prep

• Hardened the release workflow so app tags prefer curated notes from docs/releases/ and fall back to commit-generated notes only when no curated file exists.
• Scoped previous-tag lookup to app tags only, preventing plugin tags from polluting app release note ranges.
• Prepared the Windows release metadata for 0.6.1, including CLI version reporting and Groq plugin metadata alignment for plugin-groq-v1.0.2.

TypeWhisper 1.2.2 is a stable maintenance release centered on the rebuilt OpenAI integration. The bundled OpenAI plugin is now presented as OpenAI / ChatGPT and can use either a traditional API key or a ChatGPT/Codex login for prompt processing.

Highlights

• OpenAI / ChatGPT login - the bundled OpenAI plugin now supports browser-based ChatGPT/Codex OAuth in addition to API-key auth.
• Better GPT-5 and Codex support - reasoning effort can be tuned directly in plugin settings, and the bundled model list now includes current GPT-5/Codex variants such as GPT-5.3-Codex-Spark.
• Fixed Codex response handling - the ChatGPT/Codex path now speaks the actual Codex Responses protocol, including the required request shape and SSE response parsing.

Fixes & Compatibility

• Fixed 400 failures from the ChatGPT/Codex endpoint caused by the old request format.
• Improved OpenAI OAuth compatibility for the browser login flow and Codex login import.
• Restored SDK compatibility for external plugins after extending the OpenAI chat helper, so older plugin builds keep loading correctly.
• The marketplace OpenAI / ChatGPT plugin now correctly requires TypeWhisper 1.2.2 or newer.

Bug Fixes

• guard incompatible transcription engines on Intel
• disable macOS 14 settings sidebar animation
• show script/region variants in language pickers
• use curated notes for GitHub releases

Other Changes

• cover script and region language display variants
• update stable release wording
• update stable release wording in README
• rename release readiness guide
• hide legacy 1.1 filename in README

New Features

• add whisper.cpp marketplace plugin

Bug Fixes

• add new plugins to marketplace registry workflow
• clear whisper model selection on unload

Other Changes

• harden CI and prepare v0.6.0 release
• Refine settings shell and localization
• Implement custom protocol handling and improve Discord claim integration
• Add supporter Discord claims and live profile context
• Refactor structured settings pages
• Polish settings, wizard, and snippets UX
• Overhaul Windows settings UI
• Fix Win hotkey modifier handling
• Add vocabulary boosting for local Parakeet transcription

TypeWhisper 1.2 is the new stable macOS release, focused on speed, reliability, and day-to-day polish. Dictation starts noticeably faster, background CPU use is lower while idle, and audio behavior is more resilient on real-world setups like AirPods, Bluetooth devices, and changing input routes.

Highlights

• Much faster dictation start – metadata capture and browser URL resolution no longer block the critical startup path.
• Lower idle CPU usage – background listeners were tightened up to reduce unnecessary CPU activity while TypeWhisper is idle.
• More reliable audio behavior – better recovery after AirPods, Bluetooth profile switches, and AVAudioEngine reconfiguration.
• Better day-to-day usability – a new Minimal indicator, transcript preview controls, and a more guided Rules workflow make TypeWhisper easier to use.

Speed & Efficiency

• Reduced dictation start latency by moving metadata capture and browser URL resolution off the critical path.
• Reduced unnecessary background CPU usage while TypeWhisper is idle.
• Improved handling for very short clips, especially when streaming preview is active or Parakeet is selected.
• Fixed streaming-preview and no-speech-guard behavior so brief speech is retained more reliably.

Reliability

• Hardened audio engine startup and teardown around AVAudioEngine race conditions.
• Improved recovery after AirPods and Bluetooth profile switches.
• Restored the selected Apple Speech model on app launch.
• Fixed engine startup so TypeWhisper auto-selects the configured engine more reliably.
• Fixed dictation start when no microphone is available, with a cleaner failure path.
• Fixed a hotkey event tap crash.
• Improved compatibility with remapped Hyperkey shortcuts and combo hotkeys, including Caps Lock edge cases.
• Fixed WhisperKit download progress so it stays attached to the selected model.
• Hardened bundled plugin installation and update flows so installed plugin bundles remain managed correctly.

Usability & Polish

• Added the new Minimal indicator style for a compact floating status view.
• Added a toggle to disable live transcript preview for Notch and Overlay indicators.
• Reframed Profiles as Rules with a guided editor for app- and website-based matching.
• Backported the modern settings sidebar so the updated navigation also works on macOS 14.
• Added an app visibility picker to choose between menu bar, Dock, or Dock-only-while-window-open behavior.
• Polished the notch indicator layout and matched its height more closely to the display safe area.
• Fixed multiple German localization issues and other small copy and layout regressions.
• Avoided unwanted clipboard history writes when clipboard preservation is enabled.

Add-ons & Platform Updates

• Added optional HuggingFace token support to the bundled Qwen3, Granite, and Voxtral MLX plugins for higher download limits and clearer download errors.
• Added the new Gemma 4 local LLM plugin via MLX for on-device prompt processing.
• Improved Gemma 4 prompt processing and bumped the bundled Gemma 4 plugin to 1.0.1.
• Localized built-in term pack metadata in English and German.
• Added the supporter Discord claim flow for eligible supporters.

Power Users & Developers

• Added dictation session IDs to API responses for local automation workflows.
• Added GPT-5 token parameter support to the OpenAI chat helper for SDK and plugin integrations.
• Improved plugin load diagnostics with structured OSLog logging.

Bug Fix

• Fixed marketplace not showing new plugins — RegistryPlugin deserialization failed when optional fields were missing from the registry JSON. Switched from positional record to init properties with defaults.

What's New in v0.5.0

Major Architecture Change

• Replaced SQLite with JSON file storage — All data (history, dictionary, snippets, profiles, prompts) is now stored as simple JSON files. No more database migrations. Faster startup, easier debugging, and human-readable data files.
• Removed Microsoft.Data.Sqlite dependency — Smaller application footprint.

New Features

• Error Log — Structured error logging with categories (transcription, recording, prompt, plugin). View errors and export diagnostics in Settings → About.
• App-aware Formatting — Transcribed text is automatically formatted based on the target app: Markdown for Obsidian/Notion, HTML for Outlook, plain for code editors.
• Audio History — Recordings are now saved as WAV files alongside transcription history. Play back past recordings directly from the history view.
• Watch Folder — Monitor a folder for audio files and automatically transcribe them with sidecar .txt output.
• Memory Service — Automatically extract lasting personal facts from transcriptions using LLM. Stored via memory storage plugins.
• Spoken Feedback — Text-to-speech readback of transcribed text after each dictation (Settings → Advanced).
• Accessibility Announcements — Screen reader support via UI Automation live region events for Narrator, NVDA, and JAWS.
• System Audio Capture — WASAPI loopback capture for recording system audio output.
• CLI Tool — typewhisper command-line tool with status, models, and transcribe commands.
• License & Supporter Tiers — Polar.sh license activation with Bronze/Silver/Gold supporter badges.
• Release Channels — Stable, Release Candidate, and Daily update channels.
• Audio Recorder — Standalone recording feature in Settings → Recorder with auto-transcription and recording history.

UI Improvements

• Redesigned Settings Sidebar — Restructured to match Mac app: Home, General, Recording, Models, Recorder, History, Dictionary, Snippets, Profiles, Prompts, Integrations, Advanced, License, About.
• Dashboard Enhancements — Trend indicators (↑↓%) on all stat cards, "All Time" period option, recent transcriptions section.
• Overlay Redesign — Centered text layout, subtle border, larger font, more padding. Matches the new preview in Settings.
• Overlay Preview — Live preview in General settings showing actual widgets, updates in real-time when changing widget selections.
• Marketplace Grouping — Plugins are now grouped by category (Transcription Engines, LLM Providers, Memory, Actions, Utilities) with count badges.

New Plugins (14)

LLM Providers: Claude (Anthropic), OpenRouter, Cerebras, Cohere, Fireworks Transcription Engines: Gladia, Speechmatics, Soniox, Cloudflare ASR, Google Cloud STT, Qwen3 STT, Voxtral (Mistral) Memory Storage: File Memory, OpenAI Vector Memory

Plugin SDK Enhancements

• ActionCompletedEvent for tracking action plugin results
• App context (process name, window title, URL) in transcription events
• IMemoryStoragePlugin interface for memory storage backends
• IPluginSettingsActivity for progress reporting in plugin settings
• SetStreamingDisplayActive() for plugins that handle their own streaming UI
• Category and IsLocal fields in plugin manifest for marketplace grouping

Bug Fixes

• #21 — Added option to disable saving transcriptions to history (Settings → Advanced)
• #22 — Fixed intermittent hotkey toggle failures with 300ms debounce
• #24 — Hotkey events are now swallowed to prevent triggering media playback in background apps
• Hotkeys can now be cleared with Escape, Delete, or Backspace
• Timer and other widgets show proper fallback values in overlay preview

New Features

• Add vocabulary boosting to ParakeetPlugin, closes #129 (#130)
• Add V2/V3 model version selection to ParakeetPlugin, fixes #137 (#146)
• Add dynamic model fetching to GeminiPlugin, fixes #136 (#145)
• Add Google Cloud Speech-to-Text plugin (#147)
• Add dictionary JSON import and export (#148)
• Non-blocking model downloads (#149)
• Add configurable sound feedback settings, fixes #139 (#150)
• Add post-processing transparency to history and indicators, fixes #134 (#151)
• Record app audio with separate tracks, refs #141 (#152)
• Add community term pack registry (#154)
• Add Prompts & AI wizard step and cross-tab navigation, fixes #135 (#155)
• Add drag-and-drop reordering for prompts, fixes #153 (#156)
• Add media pause during recording, fixes #159 (#160, #163)
• Add mouse button shortcuts, fixes #171 (#172)
• Add Auto Enter profile setting, fixes #167 (#169)
• Add toggle to disable history saving (#168)
• Show both STT and AI-processed text in history, fixes #176 (#183)
• Sort marketplace plugins by download count, closes #186 (#187)
• Add per-profile translation toggle, fixes #199 (#201)
• Add business licensing with Polar.sh integration (#195)
• Add lifetime licenses and supporter tiers for private users (#202)
• Suppress indicator streaming text when plugin provides own display (#203)

New Plugins

• OpenRouterPlugin with model search, pricing, and credits display (#173)
• Soniox cloud transcription plugin (#174)
• Cohere cloud transcription plugin, fixes #177 (#185)
• Speechmatics cloud transcription plugin, fixes #192

Bug Fixes

• Detect Fn hotkey on release to avoid Fn+key conflicts, fixes #132 (#144)
• Update FluidAudio for Swift 6 actor safety, fixes #129 (#131)
• Fix community term pack loading state
• Fix ultrashort push-to-talk dictation (#157)
• Fix prompt toggle opening edit screen instead of toggling (#158)
• Fix notch indicator not displaying on other virtual desktops, fixes #165 (#166)
• Embed MediaRemoteAdapter framework in app bundle, fixes #159
• Fix 'No model loaded' error when engine plugin is still loading, fixes #182 (#184)
• Fix prompt toggle not responding to clicks, fixes #186 (#188)
• Fix recording started sound not playing, fixes #189 (#194)
• Fix silent recording on multi-channel audio interfaces (#197)
• Fix clipboard preservation toggle not working, fixes #198 (#200)
• Add error checking and logging for audio device selection, fixes #181 (#196)

Other Changes

• Remove App Store build support (#161)
• Disable implicit build dependencies to fix Xcode memory crashes

New Features

• add local Gemma 4 LLM plugin (LLamaSharp/GGUF)

Bug Fixes

• include NuGet dependencies and native runtimes in plugin ZIP

Other Changes

• add gemma-local to plugin publish workflow

New Features

• Silence hallucination filter — Two-layer detection (client-side energy gate + server-side no_speech_prob) prevents cloud Whisper models from hallucinating random text on silent audio (#19)
• Visual hybrid hotkey indicators — Toggle mode now shows amber overlay border and mode-aware status text ("Recording (Toggle) - press hotkey to stop") so you always know recording is active (#7)
• Plugin localization — All 6 cloud plugins (OpenAI, Groq, Gemini, Deepgram, AssemblyAI, OpenAI Compatible) now fully support English and German in their settings views (#14)
• IBM Granite Speech plugin — New local transcription engine available in the Marketplace

Bug Fixes

• App no longer freezes when no microphone is connected; shows feedback and auto-recovers when a mic is plugged in (#5)
• Groq plugin now has a 30-second HTTP timeout instead of infinite, preventing hangs on slow/unreachable API (#6)
• Prompt actions now show a clear warning when no LLM provider is configured, instead of silently doing nothing (#8)
• Translation is correctly skipped when the configured source language already matches the target language (#9)
• System language detection now uses the Win32 API directly, fixing incorrect language when launched via the Velopack installer (#15)
• Granite Speech plugin download now includes retry logic and validation (#4)

TypeWhisper 1.0 is the first stable release - a system-wide dictation app for macOS that lets you choose your own transcription engine.

Highlights

Your Engine, Your Choice Pick from local engines (WhisperKit, Parakeet, Qwen3, Granite, SpeechAnalyzer) or cloud services (Groq, OpenAI, Deepgram, AssemblyAI, Cloudflare) - all as plugins you can install and swap freely.

System-Wide Dictation Four hotkey modes (hybrid, push-to-talk, toggle, prompt palette) with a notch-integrated recording indicator. Dictation works in any app - just press your hotkey and talk.

Prompt Processing Send transcriptions through LLM providers (Apple Intelligence, Groq, OpenAI, Gemini) with bundled presets for fixing grammar, summarizing, extracting structured data, and more. Create your own custom prompt actions.

Profiles App- and URL-based profiles that automatically switch language, engine, prompt, and output format depending on where you're typing.

Plugin System A full Plugin SDK with four extension points (transcription engines, LLM providers, post-processors, action plugins). Install plugins from the built-in marketplace or build your own.

New Features

• Option to preserve clipboard content after dictation (#127)
• Reorganize Settings tabs for cleaner UX (#111)
• VoiceOver support, spoken feedback, and read-back TTS (#105)
• Error display UI with notch banner and persistent error log (#109)
• Enhance hotkey handling to distinguish modifier releases from key releases
• Inline command detection per profile, fixes #84 (#87)
• Double-tap hotkey support, fixes #106 (#107)
• Audio recorder with mic and system audio capture (#98)
• Structured data prompt presets (#95)
• Watch folder for automatic audio transcription (#99)
• CGEventTap for hotkey event suppression, fixes #101
• App-aware auto-formatting with profile-based output format, fixes #55
• Category grouping and hosting filter in plugin marketplace
• Memory plugin system with LLM extraction and correction tracking
• Kimi K2 LLM model support

Bug Fixes

• Handle Space key and special characters in hotkey recorder, fixes #94
• Resolve Cmd+V/Cmd+C on non-QWERTY keyboard layouts, fixes #92
• Improve LiveTranscriptPlugin text handling and UI (v1.0.2)
• Add confirmation dialog before clearing all memories
• Thread safety, migration, and race condition fixes

Other Changes

• Prepare TypeWhisper for 1.0 readiness (#112)
• Granular key event states, improved suppression logic, key repeat handling
• FileMemoryPlugin and OpenAIVectorMemoryPlugin added to plugin CI workflow
• German translations for all memory plugin UI strings
• Simplified memory plugin system, eliminated duplication
• 30s extraction cooldown, skip memory injection for extraction calls

For Developers

• HTTP API (/v1/*) for local automation
• typewhisper CLI embedded in app bundle
• Plugin SDK with stable contracts for 1.x
• WidgetKit widgets for stats, activity, and history

System Requirements

• macOS 14.0 or later (SpeechAnalyzer plugin requires macOS 26)
• Apple Silicon or Intel Mac

Bug Fixes

• Sparkle update detection with full git history, bump to 0.13.4

What's New

CloudflareASR Plugin

Community contribution by @yunluyl - new transcription plugin for OpenAI-compatible ASR servers behind Cloudflare tunnels.

• Supports CF-Access service token authentication and optional Bearer token
• Qwen3-ASR output parsing (<asr_text> tags)
• Model fetching and manual model selection

Audio Engine Resilience

• Recording now automatically recovers when macOS interrupts the audio engine (e.g. notification sounds, Bluetooth device changes)
• Previously buffered audio is preserved during engine restart

Menu Bar Fix

• Windows opened from the menu bar (Settings, History) now properly appear in front instead of behind other windows

Based in part on PR #88 by @yunluyl

What's New

• History Audio Playback: Optionally save and replay audio recordings from transcription history

Bug Fixes

• Cloud Transcription Reliability: Fixed cloud API requests (Groq, OpenAI, Deepgram, etc.) hanging after sleep/wake or network changes. All plugins now use ephemeral HTTP sessions with 30s timeouts instead of persistent connections that go stale
• Indicator Preview: Fixed color scheme not applying correctly in IndicatorPreviewView (#86)
• MLX Plugins: Fixed Qwen3 and Voxtral plugins failing to load on macOS 15 due to missing Swift compatibility libraries
• Plugin CI: All plugins now build with Xcode 26.2 for consistent package resolution

What's New in v0.4.1

Welcome Wizard Redesign

• Dark theme — Migrated to FluentWindow with Mica backdrop, matching the rest of the app. Previously rendered with invisible text on a white background
• Plugin-based setup — Step 1 now dynamically lists all available marketplace extensions with install/installed status and shows transcription models from installed engines. No more hardcoded model selection
• Streamlined flow — Reduced from 5 steps to 4: Extensions & Model → Microphone → Hotkeys → Done. The obsolete "Cloud Providers" step has been removed (now handled by the plugin system)
• Re-run anytime — Added "Open Wizard" shortcut on the Dashboard to re-run the setup at any time

Bug Fixes

• Fixed all Welcome Window elements being invisible on the dark theme (#3)

New Features

• Support modifier-combo hotkeys like CMD+OPT, Fn+CMD, etc. (#81)
• Expose selected text to post-processing plugins (#83)

Bug Fixes

• Fix Sparkle update detection not finding new versions (#85)
• Fix hidden history records staying selected when collapsing groups (#80)
• Fix missing plugins in registry and auto-register new plugins
• Fix plugin CI building with wrong Xcode version (#82)

What's New in v0.4.0

This is a major feature release bringing full feature parity with the Mac version.

Post-Processing Pipeline

• Refactored text processing into a priority-based pipeline architecture
• Plugin post-processors can now hook into the processing chain with custom priorities
• Action plugins (Linear, Obsidian) can receive processed text instead of inserting it

HTTP API (7 new endpoints)

• GET /v1/history — Search history with pagination
• DELETE /v1/history — Delete history entries
• GET /v1/profiles — List all profiles
• PUT /v1/profiles/toggle — Toggle profile on/off
• POST /v1/dictation/start / stop — Remote recording control
• GET /v1/dictation/status — Check recording state
• Extended /v1/status with streaming and translation capabilities

Model Auto-Unload

• Models are now automatically unloaded after a configurable idle timeout
• Saves memory when TypeWhisper is running but not actively used
• Configurable via ModelAutoUnloadSeconds setting (0 = disabled)

History Export

• Added Markdown (.md) and JSON (.json) export formats
• Export dialog now supports 4 formats: Text, CSV, Markdown, JSON

New Plugins (available in Marketplace)

• Google Gemini — LLM provider supporting Gemini 2.5 Flash, Pro, and Flash Lite
• Linear — Create Linear issues directly from transcriptions via GraphQL API
• Obsidian — Save transcriptions as Markdown notes in Obsidian vaults with YAML frontmatter
• Script Runner — Pipe transcriptions through custom shell scripts (cmd/PowerShell)
• Live Transcript — Floating semi-transparent window showing real-time transcription

Per-Profile Hotkeys

• Each profile can now have a dedicated hotkey for instant activation
• Bypasses automatic app/URL matching — press the hotkey to dictate with that specific profile
• Configure via the new hotkey recorder in the profile detail panel

Audio Device Preview

• Real-time microphone level meter in audio settings
• Visual feedback when selecting or testing microphone devices
• Preview automatically starts/stops when navigating to/from the Recording section

Database

• Schema upgraded to v6 with new columns for action plugin routing, profile hotkeys, and model tracking

New Plugins

• LiveTranscriptPlugin - Real-time live transcript overlay during recording (#77)

New Features

• Redesigned Home Dashboard as Bento Grid layout (#78)

Bug Fixes

• Fix iPhone Continuity mic causing recording to immediately stop (#76)
• Fix memory corruption crash in audio level update pipeline
• Fix WhisperKit auto-detect language detection (#75)
• Show guidance when no LLM provider plugins are installed

Updates

• WhisperKitPlugin 1.0.14

New Plugins

• ScriptPlugin - Post-process transcribed text via shell commands (e.g. tr, sed, custom scripts)
• ObsidianPlugin - Auto-export transcriptions to Obsidian vaults with daily notes support
• DeepgramPlugin - Real-time WebSocket streaming transcription via Deepgram
• AssemblyAIPlugin - WebSocket v3 streaming transcription via AssemblyAI

New Features

• Overlay indicator system with customizable styles and position presets
• Static indicator preview and style picker in General Settings
• Configurable notch display for multi-monitor setups (#65)
• Dynamic app group identifier for codesigning (PR #73)
• DEVELOPMENT_TEAM xcconfig for easier contributor setup (#67)

Bug Fixes

• Fix crash on macOS 26 when starting recording - EXC_BAD_ACCESS in StreamingHandler (PR #74)
• Fix crash when saving history records after ScriptPlugin processing
• Fix API key persistence in cloud plugins (#72)
• Fix Prompt Palette text selection and replacement in Electron apps (#70)
• Fix auto-install of plugins - users now choose in setup wizard (#66)
• Fix mic preview not restarting when switching devices in setup wizard (#63)
• Fix MACOSX_DEPLOYMENT_TARGET for all plugin targets (#64)
• Fix CODE_SIGN_IDENTITY overrides for automatic signing (#67)

What's New

Auto-Unload Models

• Configurable auto-unload - Local transcription models can now be automatically unloaded from memory after a configurable inactivity period. Choose from Immediate, 2/5/10/30/60 minutes, or Never.
• Automatic reload - When you start dictating after a model was auto-unloaded, it reloads automatically - no manual intervention needed.
• RAM savings - Frees up gigabytes of RAM when you're not actively transcribing, especially useful on 8 GB machines.

Settings: Recording > Engine > Auto-unload model

Fixes

• Bundled plugins - Fixed bundled plugins not being auto-enabled on first launch.

What's New

Instant Results with Streaming

When using Deepgram or AssemblyAI with live transcription enabled, the streaming result is now used as the final transcription. Previously, the app would discard the live result and run a redundant batch transcription after recording stopped — adding unnecessary delay. Now results appear instantly when you stop recording.

AssemblyAI Streaming Fixes

• Live transcription now works — fixed audio chunk sizing (AssemblyAI requires ≥50ms chunks)
• Multilingual support — automatically uses the multilingual streaming model for non-English languages
• No more duplicated text — fixed finality detection for the v3 streaming API

Bug Fix

• Fixed missing space in text stabilization when the transcription engine diverges mid-word

What's New

VoxtralPlugin

• Local Voxtral Mini 4B - New speech-to-text plugin for local transcription using Voxtral Mini 4B.

Setup Wizard

• Embedded Marketplace - Plugin marketplace is now embedded directly in Setup Wizard Step 2 with default engine and LLM pickers.

Homebrew

• Homebrew Tap support - Install TypeWhisper via brew install typewhisper/tap/typewhisper.

Fixes

• Control+Space hotkey - Fixed hotkey recorder not capturing modifier+key combinations like Control+Space.
• Processing spinner - Show a spinner in the notch indicator during transcription processing.
• Microphone permission - Auto-open Settings when microphone permission is missing.

What's New

Widgets

• Home Screen Widgets - 4 new WidgetKit widgets: Stats (small), Last Transcription (small), Activity Chart (medium), and History (large). Add them to your desktop for quick dictation insights.

Simulated Streaming

• Streaming for all engines - Non-streaming engines (Parakeet, Qwen3, cloud engines) now show simulated streaming text output, giving visual feedback while transcription completes.

macOS 14 Support

• macOS Sonoma compatibility - Deployment target lowered to macOS 14. All plugins now load on Sonoma as well.

Maintenance release v0.12.4

Other Changes

• Add UI localization with English default and German support

Maintenance release v0.12.1

Performance

• History section: Opens instantly — async cache loading off the UI thread and real UI virtualization (ListBox) replace the previous synchronous load that froze the UI
• Dashboard: No longer blocks app startup — history data loads asynchronously in the background
• Stats caching: TotalRecords/Words/Duration are cached and updated incrementally instead of recalculated via LINQ on every access

Other Changes

• Fix SupportedLanguages DIM test by enabling CallBase on mock
• Refresh profile transcription model list on plugin state changes
• Improve OpenAI Compatible plugin connection UX and remove inline styles
• Migrate all settings sections to WPF-UI controls
• Migrate PluginsSection to WPF-UI with CardExpander and fix expand state
• Add WPF-UI Fluent Design and migrate SettingsWindow to FluentWindow
• Add OpenAI Compatible plugin for local LLM servers
• Extract local models into SherpaOnnx plugin and unify Models UI
• Refresh plugin and model lists live after marketplace install/uninstall
• Add plugin system parity: IActionPlugin, streaming, marketplace, and localization
• Add prompt palette for AI-powered text processing via hotkey

What's New

Plugin Marketplace

Browse, install, update, and uninstall plugins directly from Settings > Integrations. Plugins are downloaded from GitHub Releases with progress feedback. Available/Installed tabs, update badges, and "Install from File" for local bundles.

Qwen3 ASR Engine

New local transcription engine powered by MLX on Apple Silicon. 30 languages, no API key required. Available as a plugin from the marketplace.

All Engines are Plugins

WhisperKit, Parakeet, SpeechAnalyzer, Groq, and OpenAI transcription engines have been migrated to the plugin system. The Models tab has been removed - each engine now manages its models in its own settings within the Integrations tab.

Hide Menu Bar Icon

New option in Settings > General > Appearance to hide the menu bar icon. When hidden, a Dock icon appears and the settings window opens on launch.

Plugin Localization

Plugins now support localized names and descriptions. German translations included for all bundled plugins and marketplace listings.

Improvements

• Engine and model picker moved to Recording settings for easier access
• History view redesigned with wider settings window and cached computed properties
• Plugins report supported languages for better profile matching
• Model download progress shown immediately for all engines
• Plugin auto-restores its model on app launch
• Settings window forwards plugin state changes for live UI updates

Fixes

• WhisperKit model download progress polling
• Prevent re-installing plugins after user uninstalls them
• SpeechAnalyzer settings view loads models on appear
• Plugin deactivation falls back to first ready engine
• WhisperKitConfig downloadBase set correctly to prevent Documents permission prompt

Simplification Release

This release focuses on removing complexity and cleaning up features that didn't carry their weight.

Removed

• Whisper mode (microphone gain boost) - removed for simplicity
• Silence detection - never worked reliably enough to ship
• Default prompt setting - unused, prompts are better managed per-profile
• Auto-submit in profiles - removed to reduce profile complexity
• Notch indicator extras - clock and battery removed, blue audio glow removed
• Preset labels from prompt actions list
• Dead code - toast overlay, hotkey migration, unused settings

Improved

• Settings reorganized - "Dictation" tab renamed to "Recording", audio sections regrouped
• Hotkey settings - more compact layout
• DictationViewModel - refactored with shared helpers, DRY event processing
• README - updated to reflect current feature set

Maintenance release v0.10.1

New Features

• Dynamic Island overlay: Configurable floating overlay with widgets (profile, model, audio level) and multi-monitor support
• Dictionary page redesign: Search bar with filtering, rich empty states with icons, segmented type selector, colored accent stripes on entry cards, polished add form and edit overlay

Improvements

• Profile widget: Show "Kein Profil" placeholder when no profile is active
• Profile settings: Hide detail panel when no profile is selected

What's New

History Overhaul

• Standalone History window - Access history directly from the tray menu, independent of Settings
• Timeline grouping - Entries grouped by Today, Yesterday, Last 7 Days, etc.
• Filters - Filter by app and time range with native picker controls
• Bulk delete - Multi-select entries and delete via context menu
• "Delete All Visible" - Delete all entries matching current filters at once
• Auto-retention - Configure automatic cleanup (keep last 30/90/180/365 days) in Settings > Advanced
• Multi-select export - Export multiple entries as Markdown, Plain Text, or JSON

Profiles

• Per-profile hotkeys - Assign dedicated hotkeys to individual profiles
• Auto-submit option - Automatically submit text in chat apps (press Enter after insertion)

API & Integration

• HTTP API endpoints for history and profiles (#30)

Improvements

• Separate Dev and Production builds with different bundle IDs and data directories
• Disable streaming preview when cloud engine is selected
• Fix translation language normalization
• Windows no longer force-float above other apps

What's New

HTTP API for External Integrations

TypeWhisper now exposes a comprehensive HTTP API for controlling the app from external tools like Raycast, Alfred, or custom scripts.

New endpoints:

• GET /v1/history - Search and browse transcription history (with q, limit, offset params)
• DELETE /v1/history?id=<uuid> - Delete a history entry
• GET /v1/profiles - List all profiles
• PUT /v1/profiles/toggle?id=<uuid> - Enable/disable a profile
• POST /v1/dictation/start - Start voice recording
• POST /v1/dictation/stop - Stop voice recording
• GET /v1/dictation/status - Check if currently recording

All date fields now use ISO 8601 format. CORS headers support PUT and DELETE methods.

Separate Dev and Production Builds

Debug and Release builds now use different bundle IDs and data directories, so you can run both side-by-side without conflicts:

• Production: com.typewhisper.mac / ~/Library/Application Support/TypeWhisper/
• Development: com.typewhisper.mac.dev / ~/Library/Application Support/TypeWhisper-Dev/

Other

• Updated README with screenshots for all settings tabs
• Added screenshot automation script

What's New

ActionPlugin System

• New ActionPlugin protocol for routing LLM output to custom actions instead of text insertion
• Bundled LinearPlugin: Create Linear issues directly from voice dictation

Bug Fixes

• Fix auto-paste not working in some apps (#29) - Text insertion now works reliably in all apps (VS Code, Notes, etc.) by removing an overly restrictive accessibility check

Bug Fixes

• Fix SmartScreen warnings for plugin DLLs on startup: Remove "Mark of the Web" (Zone.Identifier) from plugin files before loading assemblies. Prevents Windows SmartScreen balloon notifications for built-in plugins (e.g. Groq) after install or update via Velopack.

Bug Fixes

• Fix built-in plugins missing after install/update: OpenAI, Groq, and Webhook plugins are now included in the Velopack installer package. Previously they were only present in build output but not in the publish directory.
• Fix settings lost on updates: Settings persistence is now resilient against corruption with atomic writes, automatic backup before each save, and recovery from backup if the primary file is corrupt.

Bugfixes

Auto-Update: Balloon-Klick funktioniert jetzt

• Klick auf die Tray-Notification "Update verfügbar" startet jetzt direkt den Download und die Installation
• Vorher wurde der Klick ignoriert, obwohl der Text "Klicken Sie hier zum Aktualisieren" anzeigte

Auto-Start: Funktioniert nach Updates zuverlässig

• --minimized Flag wird jetzt korrekt ausgewertet
• Nach Velopack-Updates wird der Auto-Start Registry-Eintrag automatisch mit dem neuen Exe-Pfad aktualisiert
• Onboarding-Fenster wird bei minimiertem Start übersprungen
• Logging in startup.log für einfacheres Debugging

What's New

Multi-Webhook System (Mac-Parität)

• Multiple Webhooks mit individueller Konfiguration (Name, URL, HTTP-Methode)
• Profil-Filter pro Webhook (nur bei bestimmten Profilen auslösen)
• Delivery Log mit den letzten 20 Zustellungen (Erfolg/Fehler)
• Automatischer Retry nach 5 Sekunden bei Fehlern
• Neuer Bearbeitungs-Dialog für Webhook-Konfiguration

Laufende Apps im Profil-Editor

• Apps-Feld zeigt jetzt laufende Anwendungen als klickbare Vorschläge
• Ein Klick fügt die App zum Profil-Filter hinzu
• Liste aktualisiert sich automatisch

SDK

• TranscriptionCompletedEvent enthält jetzt ProfileName

Bug Fixes

• Fix rapid back-to-back dictation hanging the app - When starting a new recording immediately after the previous one finished, the app could become unresponsive. The inserting-state reset timer now gets properly cancelled when a new recording begins.

Cleanup

• Removed unused "Always Paste" option from profiles

What's New in v0.9

Plugin System

TypeWhisper is now extensible through plugins. A new plugin SDK (TypeWhisperPluginSDK) enables developers to build custom integrations:

• Extension points: LLM providers, transcription engines, and post-processors can all be added via plugins
• Typed event bus: Plugins can observe dictation lifecycle events (recording start/stop, transcription complete, text inserted, etc.)
• Host services API: Plugins get scoped keychain access, UserDefaults, data directories, and profile name resolution
• Priority-based post-processing pipeline: Multiple post-processors run in defined order

Cloud Providers are now Plugins

Groq, OpenAI, and Gemini are now bundled as plugins instead of hardcoded providers. This makes them easier to update independently and serves as reference implementations for plugin developers.

• Gemini support - New! Use Google's Gemini models for prompt processing via OpenAI-compatible API
• Dual-purpose plugins - Groq and OpenAI plugins provide both transcription and LLM capabilities
• User-installed plugins in ~/Library/Application Support/TypeWhisper/Plugins/ take priority over built-in ones

Webhook Plugin

A new built-in webhook plugin lets you send transcription results to external services:

• Configurable webhook URLs with per-profile filtering
• Automatic retry logic and delivery log

Bug Fixes

• Fixed media pause launching Apple Music when nothing is playing (#25)
• Fixed WhisperKit model disappearing from picker after first transcription

What's Changed

• Prompt Palette redesigned: Moved from notch overlay to a standalone floating panel for better usability and more screen space
• Project roadmap added: Added a roadmap document outlining planned features and development direction

What's Changed

• Add global default prompt setting (#17)
• Fix audio end being cut off on quick stop
• Move WhisperKit model storage from Documents to Application Support
• Add silence-pause mode for continuous dictation
• Remove unused DictationOverlay files

Other Changes

• Fix Velopack update discovery with per-architecture channels

Other Changes

• Add version display and update check to Info settings section
• Remove unused XAML namespace and explicit DataType from DictionarySection

What's Changed

• Add Custom Prompts feature with LLM text processing by @SeoFood in #18

Notch Indicator

New recording indicator that visually extends the MacBook notch - inspired by boring.notch.

New Features

• Notch-extending overlay: Three-zone layout (left | notch | right) with smooth Bezier ear curves, centered over the hardware notch
• Audio-reactive blue glow: Blue shimmer emanates from the notch shape, intensity reacts to voice level
• Configurable content: Both left and right sides independently configurable with 6 options: Recording Indicator, Timer, Waveform, Clock, Battery, None
• Expandable streaming text: Panel expands downward to show live transcription text with auto-scroll
• Non-notch support: Falls back to a compact centered bar on external displays without a notch
• Full buffer streaming: Removed 28-second sliding window limit - streaming preview now covers the entire recording

Changes

• Classic overlay removed (replaced by notch indicator)
• Display mode setting removed (simplified to visibility + left/right content pickers)
• 65 missing German translations added

Defaults

• Visibility: During activity
• Left: Timer
• Right: Waveform

Other Changes

• Non-blocking transcription pipeline with Channel<T> job queue

Other Changes

• Add cloud transcription providers (Groq, OpenAI) with provider architecture

What's Changed

• feat: support modifier keys (Command, Shift, Option, Control) as single-key hotkeys by @jccrump in #14

New Contributors

• @jccrump made their first contribution in #14

Other Changes

• Fix release pipeline: apply RID only to publish step
• Build separate release artifacts for win-x64 and win-arm64

What's Changed

• Add project README with features, build instructions, architecture overview, and model recommendations
• Fix welcome wizard: replace outdated Whisper model list with Parakeet TDT 0.6B and Canary 180M Flash

Other Changes

• Add release pipeline with GitHub Actions and Velopack
• Redesign dashboard with activity chart and stats
• Fix dispose guards and missing connection opens
• Add DB migration for tags column and consume trailing punctuation in snippets
• Enhance snippets: tags, clipboard placeholder, import/export, editor dialog
• Add dictionary feature: cards UI, term packs, auto-learning
• Extend sound durations and remove stop sound
• Add sound feedback with NAudio and fix playback order
• Remove Whisper reference from info section
• Add audio ducking and media pause during recording
• Redesign history as timeline with inline-expandable cards
• Add .agents/ and .claude/ to gitignore
• Enhance overlay UI with partial transcription, badges, and timer
• Add live transcription with Silero VAD for real-time partial results
• Refactor hotkeys: support hybrid, toggle-only, and hold-only modes
• Remove Whisper engine, consolidate to Parakeet-only
• Add app icon for tray, taskbar, and exe from Mac icon set
• Add browser URL detection via Windows UI Automation for profile matching
• Initial commit: TypeWhisper Windows with local Opus-MT translation