•Choose how the VoxBee icon looks in the Dock and menu bar — Auto, Light, or Dark. The dark treatment renders a charcoal chip with a gradient waveform, and Auto follows your system appearance and updates live when you switch between light and dark mode. It's a separate setting from your accent color, so you can darken the icon without dimming the in-app theme.
•On-device spoken-form normalization — when you dictate, VoxBee now writes out money, email addresses, times, and dates the way you'd type them ("twenty dollars" → "$20", "jane at gmail dot com" → "jane@gmail.com"). It's conservative by design and runs entirely offline, so it only converts when it's confident and never mangles your prose.
Changed
•Bug reports now ask for your email so we can follow up, and let you optionally attach recent app logs to help us diagnose the issue. Logs are aggressively redacted on-device — transcribed text, file names, and API keys are stripped — and you see the exact redacted text in a consent sheet before anything is sent.
Fixed
•Trial users with a paid license can now activate their key. The Settings license panel is state-aware (Licensed vs. Trial) with an inline activation field, instead of always showing a fixed "Activated" badge with nowhere to enter a key.
•License activation errors now show the real reason — a revoked, disabled, or device-limited key each gets its own message — instead of every failure being labeled "device limit reached."
Removed
•Removed the experimental Apple Foundation Models auto-format option. It added up to 3 seconds of latency to live dictation and frequently refused or garbled output, so it's been pulled while we explore a faster, more reliable formatter.
v0.9.1
Silent Auto-Update
Changed
•VoxBee now updates itself silently — new versions download in the background and install the next time you quit, so you're always on the latest build without interrupting your work. It's on by default and can be turned off in Settings, and a manual update check still shows the usual progress and What's New.
v0.9.0
Give Your Coding Agent a Voice
Added
•Model Context Protocol (MCP) support, so coding agents like Claude Code, Codex, and Cursor can call VoxBee as native tools — transcribe an audio file, speak text aloud, or stop playback. The new API tab gives you a one-line connect command per agent with your token already filled in, plus example prompts to try. Tell your agent to "summarize this PDF and read me the summary" and it does, out loud — agent-played audio routes through the floating playback dock, so you can pause or stop it from any page.
•A local API server that exposes VoxBee's transcription and text-to-speech over HTTP — OpenAI-compatible routes (so any tool built for the OpenAI SDK works by just changing the base URL to http://127.0.0.1) plus VoxBee-native routes for listing your installed models and voices. Everything runs on-device: no cloud round-trip and no per-minute billing. The API tab toggles the server, manages your bearer token, and embeds an interactive explorer for trying every endpoint in the app.
Fixed
•Changing playback speed mid-sentence no longer restarts text-to-speech from the top or skips ahead a step — playback now continues smoothly from where you were at the new speed.
•The System (on-device) voice no longer stops playing when you change speed mid-playback.
v0.8.5
Screenshot Dictation Without Speech
Fixed
•Capturing a screenshot without speaking no longer throws an error. When a recording has no usable speech but a screenshot was taken, VoxBee now hands you the screenshot instead of showing "No speech detected" or "Recording too short" — pasting the image, dropping the file path into your terminal, or leaving it on the clipboard, depending on where you're working.
v0.8.4
Notch-Anchored Recording Overlay
Changed
•The recording overlay now hugs the MacBook notch like a Dynamic Island — status text on the left, a live waveform on the right, expanding sideways from the notch as you speak. Its near-black fill blends into the physical notch, and clicks pass straight through to the menu bar. On displays without a notch, it falls back to a floating pill anchored at the top of the screen.
v0.8.2
Push-to-Talk Upgrade Fix
Fixed
•Upgrading from v0.7 no longer silently changes your push-to-talk hotkey. If you kept the default Option trigger, VoxBee now keeps Option for push-to-talk (and Option+Shift for hands-free) after the update instead of switching you to a different combo. You can still reassign your hotkey anytime in Settings.
v0.8.1
Bluetooth Dictation Reliability
Fixed
•Bluetooth headsets no longer cut dictation short when the audio route or codec changes mid-recording. VoxBee now transparently restarts the capture engine on configuration changes (codec switches, sample-rate flips, route changes), preserves the in-flight audio buffer, and keeps recording instead of bailing out.
•Releasing the push-to-talk hotkey during Bluetooth warm-up no longer leaves dictation stuck. If you tap and release while the engine is still spinning up, VoxBee now tears down the freshly-started engine cleanly instead of waiting for the 60s force-reset.
Changed
•Cold Bluetooth starts now show a brief "preparing" overlay after 250ms so the 3-4 second SCO/HFP handshake feels intentional instead of frozen. Wired mics and warm Bluetooth engines stay flicker-free.
v0.8.0
Text-to-Speech & Audio Studio
Added
•Text-to-speech across the app — a floating playback dock reads notes, meeting summaries, meeting transcripts, file transcripts, and history rows aloud. Six engines: System (free, on-device via AVSpeechSynthesizer), OpenAI, ElevenLabs, Deepgram Aura, Groq PlayAI, and Cartesia. Bring your own API key for cloud engines; audio is cached on disk by content + voice + speed, so replays never re-bill.
•Audio Studio — paste any text, pick a voice, preview, and export to MP3 or WAV. Estimated cost is shown before you commit, and exports above $1 ask for confirmation before consuming API credits.
•Configurable push-to-talk hotkey — pick any 1–2 modifier combo from Option, Control, Fn, and Shift, plus an optional hands-free escalator (a second modifier or any A–Z letter). Default for new installs is Option+Shift; existing users are migrated from the legacy single-modifier setting on first launch.
Changed
•Unified provider API keys — STT, summarization, and TTS for the same provider now share a single Keychain entry. Drop in your OpenAI key once and it powers Whisper STT, GPT summaries, and TTS without re-entry. Existing keys are migrated automatically; on conflict, the STT slot wins and the LLM key is validated in the background.
Fixed
•Mid-playback voice and speed changes in the TTS dock now restart cleanly instead of stalling, and the Transcribe tab's inline ▶ button is back.
v0.7.1
Cmd+K Search
Added
•Cmd+K command palette — press ⌘K from anywhere to fuzzy-search across history, notes, meetings, and files, and jump straight to any result.
•Dictation audio retention with replay and reprocess — push-to-talk clips are kept on-device for 7 days so you can replay them from History or re-run them through a different transcription model. Opt-out toggle lives in Settings.
v0.7.0
Google Drive Sync
Added
•Google Drive sync for meetings, transcriptions, and notes — connect once from AI & Summaries settings and VoxBee mirrors your local library to a folder it creates in your own Drive. Scope is limited to files this app creates (drive.file), syncs run automatically after recording, editing, re-transcribing, or saving notes, and you can drag the VoxBee folder anywhere in your Drive without breaking sync.
•Horizontal swipe navigation with page dots on the What's New sheet — drag between feature cards instead of scrolling.
Fixed
•Long Ollama summaries no longer fail with a 60s timeout. Local Ollama now streams tokens with a per-token idle window, so cold model loads and slow generations on big meetings complete instead of being cut off mid-summary. Cloud providers keep their fast-fail 60s timeout.
v0.6.6
Your Dictation Year at a Glance
Added
•GitHub-style usage graph on the home tab — a 52-week heatmap of daily dictation activity, with cell intensity driven by word count. Page back through prior 52-week windows with the ◀ ▶ buttons.
•Blinking red dot on the menu bar icon whenever a meeting is being recorded, so it's always clear when the mic and system audio are being captured.
Fixed
•Push-to-talk no longer gets stuck "recording" when Mission Control swallows the modifier-up event for Ctrl+Arrow space-switching chords — the hotkey manager now reconciles modifier state every 250ms and recovers automatically.
v0.6.5
Inline Re-Transcribe & Re-Analyze
Added
•Re-transcribe and Re-analyze buttons on the transcription detail header — no more right-click required to re-run transcription or LLM analysis. The right-click menu still offers the granular Re-transcribe as / Re-analyze as / Remove options.
•Pointing-hand cursor on action buttons across Transcribe (Re-transcribe, Re-analyze, Identify Speakers, Copy, Export) and Meetings (Re-summarize, Re-transcribe, Identify Speakers, Copy summary) for clearer affordance.
v0.6.4
Apple Intelligence Diagnostics
Changed
•Auto-format provider row now surfaces the specific Apple Intelligence availability reason (Preparing, Setup needed, Unavailable) instead of a generic "Setup needed" — so you can see exactly why on-device auto-format is gated off on a given Mac.
•Logs the Foundation Models availability case once at app launch to help diagnose installs where auto-format mysteriously stays disabled.
v0.6.3
On-Device Auto-Format & Summary Reliability
Added
•On-device auto-format for dictation (Beta) — on macOS 26 with Apple Intelligence enabled, Apple Foundation Models add punctuation, capitalization, and detect list cues locally before text is injected. Off by default; enable from Speech settings.
•X and LinkedIn brand links in the sidebar below Settings.
Fixed
•Summary reliability — Ollama is now preflight-checked before a run, long transcripts are chunked so they fit provider context windows, and the chosen summary model is remembered per provider instead of being reset when switching.
•Auto-format now rejects all-caps output from Apple Foundation Models and falls back to the raw transcription, so dictation never lands in shouty text.
•Auto-format no longer holds dictation hostage when the on-device provider stalls — the timeout now releases the caller immediately and falls back to raw text.
v0.6.2
More Cloud Providers & Sharper Summaries
Added
•Five new cloud speech-to-text providers — Mistral Voxtral, Cohere Transcribe, Speechmatics, Alibaba Qwen3-ASR, and Soniox. Bring your own API key from Cloud Speech settings; the purple cloud badge stays visible whenever audio leaves the device.
Changed
•Meeting summary prompt sharpened to favor concrete details — names, numbers, decisions, and action items — instead of vague recaps.
v0.6.1
xAI Grok Speech-to-Text
Added
•xAI Grok (grok-stt) as a new cloud speech-to-text provider — bring your own xAI API key from Cloud Speech settings to dictate through Grok's hosted endpoint, with smart formatting enabled by default.
v0.6.0
Cloud Speech Models & Reliability
Added
•Cloud speech-to-text providers — pick from OpenAI (gpt-4o-transcribe, whisper-1, plus a diarized variant), Deepgram (nova-3), AssemblyAI (universal-3-pro), ElevenLabs (scribe_v2), and Groq (whisper-large-v3 / turbo). Bring your own API key from AI & Speech settings; a purple cloud badge stays visible whenever audio is leaving the device.
•Fail-fast preflight before every recording — missing transcription models, summarization API keys, mic permission, and screen recording permission are now caught up front with one-click fixes (Open Settings, Choose Model, Open API Key page) instead of failing mid-pipeline.
•Structured, actionable error banners across meetings and file transcription — invalid API key, out-of-credits, model-not-found, and offline Ollama all surface human messages with primary and secondary recovery actions.
•Direct "Get API Key" links in Cloud Speech settings that jump to each provider's API key dashboard.
Changed
•Meeting pipeline telemetry now emits paired started/completed (or failed) events for transcription, diarization, and summary stages, so dashboards can detect hangs, sleep/wake stalls, and crashes between stages.
•Cloud transcription requests retry on rate limits, server errors, and transient network failures with exponential backoff (up to 3 attempts), while user-actionable errors skip retry and surface immediately.
v0.5.7
Meeting Error UX & Transcript Polish
Added
•Stage-checkpointed meeting pipeline — transcription, diarization, and summary progress now survive app crashes and relaunches, resuming at the first incomplete step instead of starting over.
•Actionable error banners for meeting summaries — missing API keys, out-of-credits, invalid models, and offline Ollama now surface a human message with one-click retry or open-settings, replacing raw API error text.
•Meeting summary retry and transcript timeline view, so you can re-run a failed summary or jump through the transcript by speaker turn.
•Clipboard is now restored after dictation in Electron apps and browsers where Accessibility detection is inconclusive — your previous clipboard content comes back instead of being replaced by the transcription.
v0.5.6
Fewer Dictionary False Positives
Fixed
•Personal Dictionary corrections are now much less aggressive on common English words — a new phonetic validation step plus consonant-group-normalized edit distance prevents words like "open" from being rewritten to brand-name dictionary entries.
v0.5.5
Smarter Text Injection
Fixed
•Text injection now always fires Cmd+V paste regardless of text field detection, preventing silent failures in Electron apps and web inputs
•Tri-state text field detection — clipboard is restored only when a text field is confirmed, and a helpful toast appears only when no text field is detected
v0.5.4
OpenRouter & Clipboard Safety Net
Added
•OpenRouter as an LLM API provider for meeting summaries and note transformations
•Transcribed text now stays on the clipboard when the paste target has no text field, so you can manually paste it
v0.5.3
Meeting Transcripts & Speaker Diarization
Added
•Speaker diarization for meetings (Beta) — on-device speaker detection identifies who said what using NVIDIA Sortformer
•Full meeting transcript display with speaker-labeled segments
•Word wrap and copy button on transcription history rows
•In-app review prompts to help us improve VoxBee
Fixed
•Replaced misleading Export button with copy feedback in the meetings tab
•Fixed a crash when stopping audio recording if no audio tap was installed
v0.5.2
Parakeet Recommended Tags
Changed
•Parakeet models are now prioritized in the model picker with a recommended tag
v0.5.0
Parakeet Models & Guided Language Setup
Added
•NVIDIA Parakeet speech models via FluidAudio, including fast English models and a multilingual European model option
•Acoustic vocabulary biasing for Parakeet live dictation so personal dictionary terms are recognized more reliably
•Guided language picker with compatible model recommendations, download flow, and mismatch warnings when the active model cannot serve the selected language
Fixed
•Recording now falls back to the built-in microphone if the default input device hangs after sleep/wake or Bluetooth audio issues
v0.4.1
Overlay Feedback & Smart Paste Fix
Added
•Error feedback on recording overlay when transcription returns empty results
Fixed
•Screenshot smart paste now works correctly in Codex and other rich-content editors
v0.4.0
Notes, Screenshots & What's New
Added
•Notes tab — voice-driven scratch pad with LLM template transformations for reformatting, summarizing, and editing dictated text
•Hands-free screenshot capture during dictation — capture screen regions without interrupting your recording
•What's New sheet shown once after each app update with release highlights
•Drop zone in media transcription is now clickable to open the file picker directly
Fixed
•What's New video now renders correctly in the looping video player
v0.3.10
UI Redesign & Download Reliability
Changed
•Redesigned UI with refined color palette, glass-effect cards, and polished interactions throughout the app
Fixed
•Robust first-time model download experience with clear state machine transitions and better error recovery
•Screen Recording permission banner now shown when meeting auto-detect is enabled
v0.3.9
Permissions & Recovery
Added
•Missing permissions banner on all main pages when microphone, accessibility, or screen recording access is needed
•App version number displayed in the About settings tab
Fixed
•Transcriptions and meetings now recovered from disk on fresh install or data reset
•Microphone permission prompt now correctly opens System Settings when previously denied
•Hotkey monitors now properly reinstall after accessibility permission is granted
•Stale accessibility permission polling no longer continues after permissions are granted
•Auto-update downloads no longer serve stale cached versions
v0.3.8
Automatic Meeting Detection
Added
•Automatic meeting detection — VoxBee detects when you join Zoom, Teams, FaceTime, Google Meet, or Webex and offers to record
•Scratchpad and collapsible history sections restored on the Home page
Changed
•History page now lazy-loads by day groups for faster performance
•Meetings feature is no longer in beta
Fixed
•Reduced false positives in meeting detection for browser-based calls
v0.3.7
Control Key Hotkey & License Fix
Added
•Control (⌃) as a third hotkey modifier option — hold Control to dictate, Control+Shift for hands-free mode
•Reduced unnecessary keychain writes that could trigger repeated system prompts
v0.3.6
Download Reliability
Changed
•App downloads now served from a dedicated custom domain for improved reliability
v0.3.5
Activation Help & Keychain Fix
Added
•Support email links in the license activation screen for easier help access
Fixed
•Fixed license access check failing after keychain migration in v0.3.4
v0.3.4
Keychain Permissions Fix
Fixed
•Eliminated repeated keychain permission prompts by migrating to Data Protection Keychain
v0.3.3
Frosted Obsidian & Reliability
Changed
•"Frosted Obsidian" visual refresh — dark glass-inspired UI across the entire app
•Faster first dictation — grammar engine now pre-warms on app launch
Added
•Text selection on transcription history and list item titles
Fixed
•Fixed license activation failing after reinstall or app update from a different build
•Fixed double-click rename not working on meeting titles
•Fixed transcriptions returning empty text when WhisperKit produces incomplete audio segments
•Fixed file transcriptions not saving immediately, risking data loss on unexpected quit
v0.3.2
Transcription Reliability & Security
Fixed
•Fixed transcription returning empty results when personal dictionary words are active — now automatically retries without dictionary biasing
•Fixed Ollama host URL validation to reject non-HTTP schemes that could leak transcript data
v0.3.1
Personal Dictionary & Settings Redesign
Added
•Personal dictionary — add custom words and aliases so VoxBee recognizes domain-specific vocabulary (names, jargon, brand terms) with phonetic matching
Changed
•Redesigned settings with a 4-tab horizontal layout (Speech, AI & Summaries, Appearance, About) replacing the single scrollable view
Fixed
•Fixed silence detection threshold rejecting normal speech as empty audio
•Fixed temp file cleanup deleting files that belong to active transcription entries
v0.3.0
Media Transcription & Intelligence
Added
•Media transcription — import audio/video files or paste a URL to transcribe with AI-powered analysis (summaries, key topics, action items)
•Automatic disk export for transcriptions and meeting summaries to ~/Documents/VoxBee/
•Update-available badges on menu bar icon, Dock icon, and dropdown menu
Changed
•Redesigned Transcribe UI with import wizard and rich summary cards
•LLM API key is no longer required for transcription — analysis gracefully skips if no key is configured
Fixed
•Fixed duplicate category label in feedback form
v0.2.1
In-App Feedback
Added
•In-app feedback form — submit bug reports and feature requests directly from VoxBee
v0.2.0
Meeting Recording & Summaries
Added
•Meeting recording — capture system audio and mic from video calls, then get an automatic AI-powered summary
•Structured meeting summaries with one-liner, discussion topics, action items, and key decisions
•Inline meeting summary editing with auto-save
•Meeting rename for easy organization
•Multiple LLM provider options for summarization (OpenAI, Anthropic, Ollama)
•Configurable summary storage folder in Settings
Fixed
•Fixed system audio capture failing silently on some devices
•Fixed silence detection in meeting recordings
•Fixed a crash when stopping audio recording due to engine teardown race condition
v0.1.17
Audio File Transcription
Added
•Audio file transcription — drag and drop MP3, WAV, AIFF, or any audio file to get a full transcript
•Smart chunking splits long recordings at natural silence boundaries for accurate results
•Timestamped transcript view with per-segment inline editing
•Export transcripts as plain text (.txt) or SRT subtitles (.srt)
•File queue with real-time progress tracking — process multiple files back-to-back
•Optional grammar correction and filler word removal for file transcriptions
•Telemetry opt-out toggle in Settings
Fixed
•Fixed waveform overlay not appearing on wired microphones
•Fixed audio latency with external devices like AirPods
•Fixed text clipping in transcript detail view
v0.1.16
Multi-Language Support
Added
•Support for 12 languages — English, Hindi, Spanish, French, German, Portuguese, Chinese, Japanese, Korean, Arabic, Russian, and Italian
•Language quick-switcher on the main screen for fast switching between languages
•Intelligent auto-switching to a compatible model when you change languages
•Model/language mismatch warning with a quick "Download Model" action
•Silence detection to prevent hallucinated transcriptions on empty recordings
•Privacy policy page on the website
Changed
•Grammar correction is now automatically disabled for non-English languages
•Accurate word counting for CJK and other non-Latin scripts
v0.1.15
Reliability & Analytics
Added
•App analytics with PostHog to help us understand usage and improve VoxBee
Fixed
•Fixed app becoming unresponsive during long transcriptions — added 60-second timeout with automatic recovery
•Fixed stuck recording state after sleep/wake or screen lock — stale hotkey presses are now detected and cleared
v0.1.14
Bug Fixes & Improvements
Fixed
•Fixed a crash in the grammar correction engine when processing long transcriptions
v0.1.13
Grammar Correction
Added
•Automatic grammar and punctuation correction powered by Harper — fixes capitalization, missing periods, and common grammar mistakes in transcriptions
•Filler word removal — automatically strips "um", "uh", "er", and word stutters from transcriptions
•Toggle grammar correction on/off in Settings
v0.1.12
Menu Bar & Dock Polish
Changed
•Cleaner menu bar layout — removed CPU icon and aligned text padding
Fixed
•Dock icon lifecycle bugs when opening and closing the main window
•Menu bar icon sizing to match macOS conventions
v0.1.11
Free Trial
Added
•14-day free trial — no account required, start using VoxBee immediately
v0.1.10
Bug Fixes & Improvements
Fixed
•Bug fixes and improvements to model loading reliability on startup
v0.1.9
License Activation
Added
•License activation system — enter your license key to unlock VoxBee
•License management in Settings with deactivate option
v0.1.8
Bug Fixes & Improvements
Fixed
•Bug fixes and improvements to update toast reliability
v0.1.7
Bug Fixes & Improvements
Fixed
•Bug fixes and improvements to update toast reliability
v0.1.6
Bug Fixes & Improvements
Fixed
•Bug fixes and improvements to update check UX
v0.1.5
In-App Update Notifications
Added
•In-app update notifications with download progress
v0.1.4
Dictation Stats
Added
•Dictation statistics on home screen — total words, duration, and average WPM
v0.1.3
Bug Fixes & Improvements
Added
•Check for Updates button in Settings and app menu
v0.1.2
Bug Fixes & Improvements
Fixed
•Bug fixes and improvements to installer and auto-update pipeline
v0.1.1
Bug Fixes & Improvements
Changed
•Default Whisper model changed to small.en for better accuracy