Help us keep the list up to date and submit new video software here.
Tool
Complete Version history / Release notes / Changelog / What's New for Vibe
v3.2.2
September 5, 2026
All releases
Your phones stay paired, and the app speaks your language everywhere
New
Phones stay paired Settings Phone lists every phone you have connected, with the name it gave and the date it paired. Each one holds its own credential, so revoking a phone you lost leaves the others connected. Revoke all is there when you would rather start over, and pairing returns you to the Phone page instead of leaving you on a code
Release notes and docs in your language every release since 0.0.2, and the Install and Models pages, are translated into all 14 website languages. An entry nobody has translated yet stays English rather than mixing two languages in one paragraph
What's new the version at the bottom of Settings opens the notes for the build you are running
Improved
Settings read better clearer section headings, and Whisper options, audio processing and phone pairing each get their own page instead of a dialog stacked on a dialog. Moving between sections is instant
The app itself is fully translated 123 strings that had fallen back to English in all 21 languages are filled in, from auto-export to the AI Assistant
Dictation output clearer icons for where the text goes, and registering a shortcut now tells you whether it worked
Right-to-left pages Hebrew reads right to left throughout, while code blocks and file paths stay left to right where they belong
Fixed
Stale shortcuts a global shortcut left registered by an earlier run is cleared at startup, so the key you press is the one that fires
Phone transcription Silero VAD resolves properly for recordings sent from a phone
Decoder limits capped at eight in both the interface and the engine, so a high value can no longer stall a transcription
v3.2.1
September 4, 2026
Intel Macs older than 2013 transcribe again
New
CPU instruction set Settings Advanced Processor lets you pick Automatic, AVX2, or Baseline (AVX) for the engine's CPU build. Automatic follows what the processor supports; Baseline is the one to pick if transcription ever dies with an illegal instruction
Fixed
Illegal instruction on older Intel Macs the Intel macOS engine was compiled for the CPU of the machine that built it, so an Intel Mac older than Haswell (2013) crashed the moment the model loaded. Intel macOS now ships the same two CPU builds as Linux and Windows and picks the right one at startup. Should the pick ever be wrong, Vibe relaunches the engine on the baseline build by itself and remembers that. Thanks to @I-G-1-1 for the report and the sysctl output (#1499)
AVX2 line in bug reports on Intel Macs it was always true; it is a real check now
v3.2.0
September 4, 2026
AI Assistant: summaries that stream, and a transcript you can ask
New
AI Assistant Settings AI Assistant sets the connection once (Ollama, OpenAI-compatible, or Claude) with a Test button that says in one line whether the key and server work. Summaries and dictation clean-up are now separate tasks, each with its own switch, presets, and prompt page, so tuning your meeting-notes prompt never changes what dictation does to a sentence
Ask about this transcript a box under the summary answers questions from the transcript itself ("What was decided about the budget?"), streams the answer, and keeps the thread in the project so it is still there next month
Prompt presets and placeholders start from Summary, Meeting notes, TL;DR, Translate, or Rewrite as article, and use {transcript}, {language}, and {speakers} in your own prompt. Dictation has Punctuation only, Formal, and Casual, with a Try it box to see what a sentence becomes before trusting it with real words
Improved
Summaries stream in the text appears as the model writes it, and the Summary tab opens itself when a summary starts
Long recordings a transcript too long for the model's context is summarized in parts and then as a whole, with a part counter, instead of being cut off
Dictation clean-up runs its own prompt, and if the model fails the raw words are still typed or copied
Fixed
Dictation ran the summary prompt dictated text used to be sent through whatever the summary prompt said, which turned a sentence into meeting notes. Each task now has its own prompt
Settings carry over the old AI settings (key, model, prompt) are migrated on first launch; nothing needs to be entered again
v3.1.11
September 4, 2026
Auto-export: every transcript written where you want it
New
Auto-export turn it on once in Settings Transcription and every transcript is written as soon as it finishes, in the formats you pick (VTT, SRT, text, Markdown, JSON, CSV, DOCX, PDF, HTML), next to each recording, in the projects folder, or in one folder you choose. Existing files are kept unless you ask to replace them, and a recording folder that cannot be written falls back to the projects folder. This brings back what the old batch page did, for the whole app
The batch moment transcribing several files now shows progress across the run with a time estimate, and ends with a card that says how many were exported, skipped, or failed, with an Open folder button, plus one desktop notification for the whole batch. Each file in the list says what it produced
Export menu in a batch the Export button becomes a menu with Export this transcript, Export all finished for anything done before auto-export was on, and the auto-export setting. Thanks to @LivingInternet for the report (#1480)
Improved
Theme switch a sun and a moon in Settings instead of a dropdown
Every audio track is heard a recording with several audio tracks (OBS, Discord, one per speaker) used to be transcribed from one track only; the rest is now mixed in, so every speaker appears and the engine no longer repeats a line over silence it was never meant to hear
Fixed
Repeated sentence for minutes the loop reported on multi-track recordings, where the model heard silence on a dropped track while the speaker kept talking, is gone with the change above (#1518, #1505)
v3.1.10
September 4, 2026
Correct transcripts on AMD graphics with older drivers
Fixed
No Vulkan runtime, no problem a Windows PC without a Vulkan loader (no vulkan-1.dll, typical of machines with no GPU driver installed) could not start the transcription engine at all. The engine now opens the loader at runtime and simply runs on the CPU when there is none. Upstream: ggml-org/ggml#1614
Correct transcripts on AMD graphics with older drivers on some AMD GPUs running a 2020-era Windows driver (seen on a Ryzen laptop's Radeon iGPU), the GPU produced nonsense: a lone "!" or random characters, while the CPU was right. The driver leaves the GPU unclassified in the engine, and the attention kernel then asked for a subgroup width the hardware cannot run. Sona 0.6.5 only asks for it where the driver can grant it, and the GPU transcribes correctly again, several times faster than the CPU. Upstream fix proposed to ggml (ggml-org/ggml#1619)
v3.1.9
September 4, 2026
Transcription is back on older AMD graphics drivers
Improved
Errors now say what the engine printed when Sona dies mid-transcription, the error dialog, the logs, and the bug report carry its last lines instead of a bare "failed to read sona event line", so problems like the one above are identifiable from a report
Fixed
Transcription is back on older AMD graphics drivers since the move to the new engine, machines with a 2020-era AMD driver (Vulkan 1.2, common on Ryzen laptops with Radeon graphics) crashed at the first GPU submit with "vkQueueSubmit: Invalid queue". The driver returns an empty queue handle from vkGetDeviceQueue2; Sona 0.6.4 falls back to vkGetDeviceQueue and the GPU works again. Updating the AMD driver is still a good idea, but no longer required
Older Intel and AMD CPUs are back Vibe no longer refuses to transcribe on a CPU without AVX2. Sona ships its CPU engine twice, an AVX2 build and an AVX baseline build, and picks the one your processor can run when it starts. Any x86-64 machine from 2011 on now transcribes; the "Your CPU is not supported" message is gone
Out of GPU memory no longer kills the run when the graphics card runs out of memory partway through a file, Vibe reloads the model on the CPU, retries that file once, and finishes the rest of the queue on the CPU. That model is remembered as too large for your GPU and loads on the CPU from then on. Thanks to @mikemikimike for the report and the first pass (#1471, #1472)
v3.1.7
What's new?
Name your speakers click a "Speaker 1" label in a finished transcript and type a name. It replaces the label on every line of that speaker, on screen and in every export (text, Markdown, HTML, SRT, VTT, PDF, DOCX; JSON keeps the index and adds speakerName). Names are saved with the project, so they are there when you reopen it
Put a line on the right speaker the same popover moves one line to another speaker, or to a new one. A short line that diarization left with no speaker at all now shows a dashed placeholder so it can be given one. Sona 0.6.2 also attributes such lines to the nearest speaker turn within 1.5 seconds, so far fewer arrive unlabelled in the first place
A Whisper prompt no longer blocks Parakeet or Nemotron those engines take no prompt, and a saved one made every run fail with "does not support text prompts". The prompt is now left out of the request when the loaded model reports it cannot use it (the translate switch too), so it stays saved for when you switch back to Whisper. The Settings prompt field says so when the selected model ignores it
Transcribe several links in one go paste a batch of links into the link box, or add them one by one with the plus button, and they queue up in a list under it. They download in turn; one that fails is skipped and listed at the end, and the rest go straight to transcription. The box clears as soon as the run starts
Links with non-Latin titles no longer fail on Windows yt-dlp can print bytes that are not UTF-8, and reading its output aborted the whole download on the first one. Thanks to @mikemikimike (#1473)
Signed and notarized on macOS and Windows.
v3.1.6
What's new?
Scroll back through a transcript while it is still being written following the live text used to be impossible to escape: scrolling up, or clicking a line to play it, snapped you straight back to the tail on the next segment. The hand-off to the reader keyed off a scroll event that was ignored for 700 ms after any scroll Vibe made itself, and every arriving segment refreshed that window, so during a real transcription it never closed. Following now ends on the gestures you actually make wheel, trackpad, scrollbar, Page Up and picks up again when you scroll back to the bottom
Google Meet detection asks for the permission it needs listed in the 3.1.5 notes, but the change landed just after that tag; this is the first build that has it. Reading a browser's window title requires Screen Recording on macOS, and Vibe checked for it without ever asking. It now requests it once and links straight to the settings pane if you decline. Zoom and Teams never needed it
Signed and notarized on macOS and Windows.
v3.1.5
What's new?
PDF export is a real PDF it used to drive the system print dialog, which on macOS returned before the job spooled and handed the printer a blank page. Vibe now generates the file directly: selectable text, embedded fonts, page numbers, and pages that break on their own
Hebrew and Arabic exports read the right way round the previous PDF engine had no bidirectional text support at all: it resolved one direction per line and reversed every glyph in it, English words and numbers included, so a Hebrew transcript came out backwards. Every export is now checked against the Unicode algorithm before it ships
The export dialog remembers you format, content, theme and both switches persist instead of resetting each time you open it, and the whole panel was rebuilt around one consistent row: nine one-line formats, a light/dark choice for the file itself, and a one-off text direction override
Word documents are set for paper a real font that covers Hebrew and Arabic, a proper type scale, speaker labels that stay with their line across a page break, and no dark page. Their timestamps read forwards again
Transcribe on the CPU a new setting in Settings Models for machines whose GPU driver crashes. The engine always supported it; nothing exposed it
Meetings are detected in about a second the detector waited out a five-second debounce on every change, so a call that had already started stayed invisible for most of a sentence. Quitting a call clears the prompt promptly now too, and the window and process scans it relies on got cheaper rather than more frequent
Google Meet detection asks for the permission it needs reading a browser's window title requires Screen Recording on macOS, and Vibe checked for it without ever asking. It now requests it once and links straight to the settings pane if you decline. Zoom and Teams never needed it
The transcription engine draws the same samples whisper.cpp does sona v0.6.1, which also settles the output: the same audio now transcribes identically twice, where before it drifted. Metal on macOS, Vulkan on Linux and Windows
A long list of recordings scrolls smoothly the sidebar only renders the rows you can see, so hundreds of projects cost what ten used to
Delete every project at once Settings Transcription, under the projects folder
Translations filled in across all 21 languages, including the transcription language lists in Turkish and Swedish, which were almost entirely English
v3.1.4 Beta
What's new?
Vibe notices meetings without interrupting them opt in under Recording settings and Vibe can detect active Google Meet, Zoom, and Microsoft Teams calls from real microphone use. The compact floating prompt stays above the call, does not steal focus, and disappears automatically
Record first, transcribe when you choose recordings are saved immediately as durable projects. Automatic transcription is now optional and off by default; audio-only projects have a polished language-and-options view with a clear Transcribe action
Recording works from anywhere configure a global start/stop shortcut, choose microphone and system audio from the meeting prompt, and manage macOS recording permissions directly in Settings
Long transcripts stay smooth transcript rows and export previews are virtualized, the scrollbar is easier to find, and Jump to playing appears only while audio is actually playing
Playback speed is one click cycle through the useful speeds directly from the player instead of opening another menu
Export is visual and predictable one export flow provides live previews for PDF, DOCX, HTML, Markdown, JSON, and text formats, respects dark mode and RTL, uses sensible filenames and Downloads defaults, and reveals saved files reliably
Project titles behave like text click the title to rename it in place, with the caret exactly where you clicked and the sidebar updating while you type. Generated names now clearly identify Record, URL, and File sources without leaking media extensions
Summaries are easier to control optionally summarize new transcripts, retry or improve existing summaries, use a roomier prompt editor, and send up to 64K tokens of meeting context
Cleaner agent setup the API & Agents page focuses on installing the Vibe skill for Codex or Claude, with a short example prompt instead of cURL and starter-prompt clutter
Autostart stays out of the way launching with the system now opens Vibe hidden, including users who already had autostart enabled
Desktop translations are complete all 22 languages now have the same 574 translated keys, including the new recording, meeting, export, summary, and agent experiences
v3.1.2
What's new?
Your settings stick again every setting reverted to its default on restart in 3.1.1. The file on disk was always correct; the app simply never read it back. If you gave up re-setting your preferences, they are safe now
Failures say why a sidecar that dies now reports its exit code and signal instead of a bare "sona process died", and a crash mid-transcription reads as a crash rather than "error decoding response body". Whisper's own diagnostics reach the log again, after being silently discarded by a no-op callback
Broken model downloads are caught files are verified before they are used, incomplete ones are detected on startup and offered a re-download, and an interrupted download can no longer masquerade as an installed model
Half the memory to load a model measured 294 MB 216 MB on a small model, and roughly one model copy less on large ones. Machines that used to run out of memory mid-load have room now
A CPU without AVX2 is named instead of dying anonymously, Vibe says the processor is unsupported and why
The phone pairing QR scans one wrong character in the QR generator misplaced the alignment patterns on any code past a certain size, and the pairing code was just past it. No scanner could read it
Recordings are 9 dB louder mixing in system audio was quietly halving the level, then halving it again. Every recording made with system audio selected was far quieter than it should have been
Dictation defaults to a two-key shortcut Option+Space on macOS, Ctrl+Space elsewhere, instead of a three-key chord
yt-dlp stops nagging the update prompt appeared on every launch and forgot "Later" each time. It is now a dismissible toast at most once a week, and it offers the update when a download actually fails, which is when it helps
Install the agent skill Settings API & Agents writes it to Claude Code and Codex so it is available in every session, with the transcripts folder and the local API baked in
Windows: the VC++ runtime is version-checked the installer only asked whether any runtime was present, so machines with an old one were skipped and crashed later with no message
The product name and a mistranslated setting fixed across all 22 languages
v3.1.1
What's new?
Record from your phone: scan the QR code in Settings Phone once, and your phone can send recordings to this computer it transcribes them with the model you already have loaded and sends the text straight back
The phone side is a web app you add to your home screen: nothing to install from an app store, no account, and it works on cellular as well as wifi
Audio travels over an encrypted peer-to-peer connection built on iroh phone browsers can't hole-punch, so it passes through a relay that only ever sees ciphertext, never your audio
Phone recordings are saved to Documents/Vibe and show up in Recents like any other transcript, and a recording waits safely on the phone until your computer is reachable again
The phone's language picker comes from whichever model you have loaded, so it always offers exactly what your desktop can actually do
New "Launch at startup" setting, off by default, sitting next to "Keep running in the tray" together they keep the desktop reachable when you reach for your phone
Every new screen is translated into all 22 languages
v3.1.0
What's new?
Redesigned the whole app around one drop-to-transcribe screen, with a recents sidebar you can resize and search
Rebuilt transcript editing: click anywhere on a line and the caret lands where you clicked, Enter/Tab/arrows move between lines, Escape reverts
The line being spoken is lit as it plays, the view follows along, and a timestamp click plays from there
New view options for the transcript text size, timestamps, speaker labels and text direction
AI summaries now run from the main window: automatically after a transcription when enabled, or on demand from the transcript's AI menu, rendered as markdown and saved with the transcript
Re-transcribe from the sidebar now asks for language, model and options first
One picker takes files and folders together on macOS, and folders keep working by drag and drop everywhere
The global dictation shortcut is set by pressing the keys, with ready-made combos for the ones macOS reserves
Optional "keep running in the tray": closing the window leaves Vibe running for global dictation, off by default (thanks @eggie04)
Settings moved out of the browser store into an editable app_config.json agents and people can read and change it, and Vibe picks up the edit immediately
Every number field in settings became a stepper you can hold, type into or reset, with "Never" and "Auto" spelled out
An update waiting is shown on the sidebar toggle instead of hiding in a menu
Fixed uppercase extensions being ignored, so a folder of .MP3 files is no longer empty (thanks @plmelancon)
In-app bug reports now arrive titled with the actual error instead of "App reports bug"
Added Czech (thanks @feretCZ), Bulgarian and German (thanks @nimdassdev); every one of the 22 languages is complete
RTL fixes throughout: the sidebar stays on the window's left, the player timeline drags the right way, and shortcuts read left to right
v3.0.23
What's new?
Fixed Vibe and Sona remaining open after exiting on Windows, which could keep GPU memory occupied and prevent relaunching
Added an Advanced setting to automatically unload inactive transcription models and release RAM/VRAM; set it to 0 to keep models loaded
Protected active uploads, VAD, diarization, and long transcriptions from idle unloadingthe timeout begins only after processing finishes
Improved model reuse so Sona skips reloading an already loaded matching model
Added translations for the model inactivity setting in every supported language
Updated Sona to v0.3.5 with Windows parent-process monitoring and reliable child-process cleanup
v3.0.22
What's new?
Added Nemotron 3.5 and Parakeet TDT v3 model support, including streaming transcription for dictation
Updated Sona to improve transcription and clarified the available transcription options
Migrated the desktop app and website to Paraglide i18n and added translation tests
Improved the model selection flow
Improved the Wall of Love design
Stop Sona cleanly before installing app updates
Made Windows code signing optional and fixed build configuration
Improved website deployment and type-checking reliability
v3.0.21
What's new?
Redesigned the main window and rebuilt Settings with a cleaner sidebar layout
Added toggle mode for Global Dictationpress once to start and again to stop
Added an optional floating dictation indicator so recording status stays visible
Improved Sona startup, model loading, error reporting, and process reliability
Fixed Windows console popups when detecting GPU devices
Improved hotkey transcription errors and saved recording filenames
v3.0.20
What's new?
Sona has been completely rewritten in Rust for improved performance and maintainability
Faster startup and a more streamlined architecture
Built-in speaker diarization, no separate runner required
Simpler build and packaging process for contributors
v3.0.19
What's new?
Remember selected audio device across sessions
Improved CPU compatibility check (clear message if AVX2 is not supported)
Fix batch transcription when folders contain files without extensions
Properly stop background audio process when quitting the app
Various internal improvements and cleanup
v3.0.18
What's new?
Improve transcription failure messages to make issues easier to understand
Update the transcription engine for better stability and compatibility
Fix build issues
v3.0.17
What's new?
Add Stable Timestamps mode for much more accurate subtitle timing in videos, tutorials, movies, and TV shows (can be enabled in More Options)
v3.0.16
What's new?
Fix macOS system audio recording
Improved structured error messages for clearer failures
Internal improvements and dependency updates
v3.0.15
What's new?
Configurable recording save path. choose where your recordings are stored
Improved system audio recording on macOS for better reliability by @Ada-lave in #978
New "Wall of Love" section with international support and better mobile layout
Recent languages are now remembered for faster language switching
Easier copy actions with a new copy button for commands
Visual polish and smoother animations across the app
Better handling of edge cases like missing audio devices
v3.0.14
What's new?
Improved Windows code signing reliability (more robust and consistent signed builds)
Safer and more reliable Windows signing flow using a remote YubiKey signing server
Better Windows tooling setup and verification steps (helps avoid install and update issues)
Cleanup of unused macOS signing configuration
v3.0.13
What's new?
Polished macOS installer design
Windows app is now code-signed for better security and fewer warnings
macOS app is now code-signed for better security and fewer warnings
Better Linux compatibility (updated Ubuntu builds for wider glibc support)
Safer transcription flow now checks audio files before running
Dependency handling improved on Linux for smoother installs
Fixed YouTube downloads on ARM devices
v3.0.12
What's new?
Disable transcribe button when no model is selected with helpful hint
Fix model loading crash for users with non-ASCII usernames on Windows
Graceful fallback to CPU when Vulkan GPU drivers are missing
Better error messages when sona crashes during model loading
Download models link moved to top of settings for easier access
Sona now reports version and commit hash on startup
v3.0.11
Whats new?
Real-time audio visualizer while recording see live feedback as you speak (#950)
Re-summarize option quickly regenerate summaries with one click
Improved dictation dialog more stable global shortcut handling
Fixed Ko-fi dialog not closing properly
Small Linux UI fixes
v3.0.10
What's new?
Add optional GPU device selection
Fix segment timestamps scaling in JSON/CSV
Fix RTL layout issues
Bypass system proxy for localhost connections
v3.0.9
What's new?
Fix diarization (speaker recognition) failing on YouTube downloads and non-16kHz audio files
Fix large file upload failing with "Invalid argument" error
v3.0.8
What's new?
Speaker Diarization
High-quality speaker detection with Parakeet identifies up to 4 speakers automatically.
Global Dictation Hotkey
Hold a shortcut anywhere to record, release to transcribe. Copies to clipboard or types at cursor.
More Format Support
MXF video, M4B audiobook, and CSV export.
Flexible Summarization
Works with any OpenAPI-compatible backend.
Sona Engine Upgrade
File size limit increased from 1GB to 15GB.
v3.0.7
What's new?
Support for much larger audio files (up to 15GB, was 1GB)
More reliable model loading with automatic retry on connection hiccups
Better sona binary detection on Linux (Arch, CachyOS, AUR installs)
Clear error message when no model is selected instead of cryptic crash
Improved error diagnostics for faster issue resolution
Option to disable anonymous analytics in Settings
v3.0.6
What's new?
Major architecture upgrade: Vibe now uses a Sona sidecar instead of an in-app whisper process
More reliable transcription with a local HTTP flow (OpenAI-compatible, streamed)
Live progress + segment updates while transcribing
Better stability and isolation (fewer crashes, easier debugging)
Improved command-line support (better CLI detection + ffmpeg handling)
More stable builds across platforms (Windows + Linux fixes)
You can now see the local API address in Settings
Anonymous analytics + better error tracking (helps catch issues faster)
v3.0.5
What's new?
Add Russian language support for i18n configuration
Enhance model download functionality and set Hebrew-AI model for Hebrew locale
Replace largest Ivrit model with faster Turbo model (almost as accurate as Large v2, better at subtitle segmentation)
Update model conversion instructions for improved setup and efficiency
Smarter transcription! You can now control how the model chooses what to say (sampling strategy + beam size)
Transcribe Folder recursively
v3.0.2
What's new?
What's Changed
update ytdlp and add check for updates by @thewh1teagle in #500
v3.0.1
What's new?
Clean updater files #410
Remove file associations #478
Keep system awake while transcribe/record #337
More logging in setup
Fix pdf export colors #455
Add Vietnamese language
pdf dark mode #455
docs: Add clarity about coreml file usage by @andrewginns in #442
Feat/linux installer by @thewh1teagle in #448
feat: sign tauri plugins by @thewh1teagle in #450
Add zh-HK and fix zh-CN in static locales by @xinbenlv in #460
Add Norwegian Language by @MechanikGamer in #459
New Contributors
@andrewginns made their first contribution in #442
@xinbenlv made their first contribution in #460
@MechanikGamer made their first contribution in #459
v3.0.0
What's new?
Improved privacy policy design and user experience
Added a link to the Vibe website for privacy policy access
Enhanced Linux installation options for a smoother setup
Fixed footer and landing page design improvements
Added installation notes and documentation to help users get started easily
v2.6.5
What's new?
Ollama support for seamless integration!
Improved logging from whisper.cpp with Vulkan.
Updated whisper.cpp for enhanced performance.
Updated Tauri for the latest improvements.
Added support for exporting transcriptions in DOCX format.
Start transcription automatically right after video download.
Fixed crash issue when Documents folder is not found.
v2.6.3
What's new?
Various miscellaneous improvements for enhanced performance.
Bug report updated for better issue tracking.
Fix #333.
Fix #331 by applying default to release profile.
v2.6.2
What's new?
Update whisper.cpp and potentially fix vulkan errors
v2.6.1
What's new?
Update whisper.cpp. should fix some issues with Vulkan
v2.6.0
What's new?
Summarize with Claude API for a more efficient and intuitive experience!
v2.5.6
What's new?
Upgraded to Tauri v2 for improved performance and stability
Enhanced tooltip design for a more user-friendly experience
Improved contrast in dark mode for better readability
Word timestamps are now automatically disabled during speaker recognition
v2.5.5
What's new?
Add option to download audio from popular websites
Fix issue with diarization; ensure word timestamps are disabled if diarization is enabled
v2.5.4
What's new?
Check that Vulkan init works or provide instructions to download older vibe version
v2.5.3
What's new?
Updated Italian translations thanks @rsaleri
Include Vulkan runtime and VC++ Redistributable installer on Windows
v2.5.2
What's new?
Fetch models directly from GitHub, with no more dependence on other cloud services
v2.5.1
What's new?
Improved recording filenames for better organization
Default to storing documentation files in the docs folder
Normalize audio before transcription for improved accuracy
Added support for special languages in speical models by model filename pattern
Enhanced logging for better debugging and issue tracking
Updated Tauri and fixed various package issues
Fix(windows): show whisper.cpp errors in Windows correctly by redirecting stdout/stderr/ experimental
Fix(windows): embed vulkan runtime DLLs
Fix(linux): add vulkan runtime deb packages
Added Vulkan SDK support for older CPUs
v2.5.0
What's new?
Vulkan support for AMD, NVIDIA, and Intel GPUsno need for CUDA, with faster computing
All models available for manual install. see Pre built models
v2.4.0
What's new?
Fix language detection and preference on launch
Fix microphone recording on macOS
Added speaker recognition (diarization)
All models available for manual install. see Pre built models
v2.2.0
What's new?
Added support for Hindi (Thanks @lovishchhabra)
Enabled back OpenCL for Windows, improving performance on compatible hardware
Enhanced GPU device information and set GPU preference in settings for better control
Improved main navigation UI for a smoother user experience
Refined overall UI design for a more polished look
Removed Windows 7 support to focus on more recent versions
Fixed incorrect timestamps by using custom whisper.cpp
Contributors
@lovishchhabra
lovishchhabra
v2.1.0
What's new?
Improved internationalization support with custom locale detection
Added option to transcribe word timestamps
Enhanced macOS DMG installation background
Set GPU preference to high performance on Windows by default
Choose GPU device for improved performance (Thanks @israelxss!)
Enhanced text manipulation with 'replaceAll' feature
Added Windows portable support
Added Swagger documentation for server APIs
Improve Polish translations (Thanks for @GitesHubisz)
v2.0.6
What's new?
Fix linux i18n (Thanks for @oleole39)
Add option to transcribe word timestamps
Add macOS dmg installation background
Set GPU preference to high performance on Windows by default
Max letters per sentence! (Thanks for @sdimantsd)
Contributors
@sdimantsd
@oleole39
sdimantsd and oleole39
v2.0.5
What's new?
Speed up by caching model context instead of reload it everytime (Thanks for @Y-PLONI for suggest it!)
Improve Portuguese translations (Thanks for @josemoura212)
Add Polish translations (Thanks for @GitesHubisz)
Fix typo in Install.md (Thanks for @eltociear)
Contributors
@eltociear
@Y-PLONI
@josemoura212
@GitesHubisz
eltociear, Y-PLONI, and 2 other contributors
v2.0.4
What's new?
Catch whisper.cpp panics instead of crashing
Pretty app version in settings at bottom with nvidia / older cpu labels (Thanks for @josemoura212 for suggestion!)
Fix batch export as json (Thanks for @DArlund for reporting!)
Add deep links. you can add vibe://download?url=<any model url> link to your website for let vibe users download modesl!
Contributors
@DArlund
@josemoura212
DArlund and josemoura212
v2.0.3
What's new?
Add support for m4a (Thanks for @yairl for reporting!)
Fix tabs UI contrast (Thanks for @oleole39 for suggestion!)
Add nvidia, opencl, and rpm support for Linux (Thanks for @thegrasshopper104 for the suggestion!)
Add option to paste model link and download directly in vibe
Contributors
@yairl
@thegrasshopper104
@oleole39
yairl, thegrasshopper104, and oleole39
v2.0.2
What's new?
Add option to record from speakers / microphone! (macOS support included)
Fix audio file encoding issues
Improve French translation (Thanks for @oleole39!)
Add rpm installer
Contributors
@oleole39
oleole39
v2.0.1
What's new?
Add French translation (Thanks for @oleole39!)
Add more Chinese translation (Thanks for @Ifan24!)
Add cli support. use Vibe from console directly!
Add Nvidia v11 and v12 (Cuda versions)
Contributors
@Ifan24
@oleole39
Ifan24 and oleole39
v2.0.0
if CPU is unsupported - Show error message and open URL for fix on Windows
Enable metal framework on macOS for GPU and improve speed by ~40%
Update to whisper.cpp version 1.6.2
Add Nvidia to releases
Add release for older CPUs
v1.0.9
New formats: JSON
Fix saving path
Fix translation in tooltips
Remove focus color from print button
Better toast message on saving files
Fix timestamps for PDF / SRT
v1.0.8
New formats: html, pdf
Show toast message after save file
Option to Print pdf
Add discord button in settings
Option to translate into English (advanced options) | Thanks for hbacelar for the suggestion!
Optimize windows with openblas
More tooltips
Fix extra line in srt
v1.0.7
Fix not a number JS error in realtime preview (Thanks for @Ifan24 for reporting)
Fix when opening settings, keep default transcribe langauge instead of changing it
Imrpve design by making filename in the audioplayer as link when hovering
Show logs in Windows by execute the following in cmd.exe:
set RUST_LOG=vibe=trace
%localappdata%\vibe\vibe.exe
add logs to check if cpu supports f16c instruction
use crate showfile to open files as selected state in file manager
Add Nvidia binary (transcribe 1 hour in less than 5 minuets)
Add non avx binary in case of old CPU crash
v1.0.6
Add Chinese language (Thanks for @Ifan24)
Add batch transcribe (Thanks for @renatoianhez)
Add error boundary in case of fatal render error
Fix Windows file path open crash
Improve language input UI by using option groups
Improve audio player
Add native translation to languages selector
Contributors
@renatoianhez
@Ifan24
renatoianhez and Ifan24
v1.0.5
Add option to drop files to window
Add option to open file by right click -> Open with
file drop animation
Support more audio formats such as opus (Thanks for @renatoianhez for the suggestion)
Add crash log to panic hook
Add option to open app config dir for getting logs
Improve ux by replacing onClick with onMouseDown
Add params info on hover in advanced options (Thanks for @NHLOCAL)
Improve default window size
Improve settings window design
Add Swedish language, (Thanks for @2bbe)
Fix taskbar progress animation
Update to latest tauri core / plugins
Improve offline installation
Option to cancel model download
Contributors
@renatoianhez
@2bbe
@NHLOCAL
renatoianhez, 2bbe, and NHLOCAL
v1.0.4
Smooth window creation
Add min window width and height
v1.0.3
Fixed model customize #54
Allow open settings by click #53
Add option to select whether to play sound / focus window when transcription complete
Use custom eyre with serialize backtrace
Window maximized on first open
v1.0.2
Wider UI
v1.0.1
Fix on windows failed to load model if Username contains Hebrew characters
v1.0.0
GPU optimization for Windows with OpenCL! (x1.5 faster)
Commit hash in system info collection
Fix translation in Hebrew for cancel transcript
v0.0.9
Realtime preview of transcription
Option to abort transcription in progress
Show percentage while transcribing
Show progress when updating
v0.0.8
Add Portuguese language, thanks to josemoura212!
Fix invalid segment error due to invalid utf-8 characters returned from whisper by adding a patch to whisper-rs
v0.0.7
Better errors reports
Upgrade tauri to v2
v0.0.6
Windows shadows
Center window on open
Fix export as txt extension
v0.0.5
Auto updater
Allow export in multiple formats
vibe v0.0.4
Fix: remove model hash verify
Add modal if error happens
Add option to reset app
Improve error reporting to include log
Add MacOS support
Optimize GPU using CoreML
v0.0.3
Settings page
Transcribe parametrs (prompt, temperature, etc...)
Better bug reporting
Option to select / download another models
v0.0.2
Optimize performance by 50%using OpenBLAS
Fix multi language encoding
Add format options as srt vtt and normal
Improve design
Common model path per CLI and Desktop
v0.0.1
remove print
September 5, 2026
All releases
Your phones stay paired, and the app speaks your language everywhere
New
Phones stay paired Settings Phone lists every phone you have connected, with the name it gave and the date it paired. Each one holds its own credential, so revoking a phone you lost leaves the others connected. Revoke all is there when you would rather start over, and pairing returns you to the Phone page instead of leaving you on a code
Release notes and docs in your language every release since 0.0.2, and the Install and Models pages, are translated into all 14 website languages. An entry nobody has translated yet stays English rather than mixing two languages in one paragraph
What's new the version at the bottom of Settings opens the notes for the build you are running
Improved
Settings read better clearer section headings, and Whisper options, audio processing and phone pairing each get their own page instead of a dialog stacked on a dialog. Moving between sections is instant
The app itself is fully translated 123 strings that had fallen back to English in all 21 languages are filled in, from auto-export to the AI Assistant
Dictation output clearer icons for where the text goes, and registering a shortcut now tells you whether it worked
Right-to-left pages Hebrew reads right to left throughout, while code blocks and file paths stay left to right where they belong
Fixed
Stale shortcuts a global shortcut left registered by an earlier run is cleared at startup, so the key you press is the one that fires
Phone transcription Silero VAD resolves properly for recordings sent from a phone
Decoder limits capped at eight in both the interface and the engine, so a high value can no longer stall a transcription
v3.2.1
September 4, 2026
Intel Macs older than 2013 transcribe again
New
CPU instruction set Settings Advanced Processor lets you pick Automatic, AVX2, or Baseline (AVX) for the engine's CPU build. Automatic follows what the processor supports; Baseline is the one to pick if transcription ever dies with an illegal instruction
Fixed
Illegal instruction on older Intel Macs the Intel macOS engine was compiled for the CPU of the machine that built it, so an Intel Mac older than Haswell (2013) crashed the moment the model loaded. Intel macOS now ships the same two CPU builds as Linux and Windows and picks the right one at startup. Should the pick ever be wrong, Vibe relaunches the engine on the baseline build by itself and remembers that. Thanks to @I-G-1-1 for the report and the sysctl output (#1499)
AVX2 line in bug reports on Intel Macs it was always true; it is a real check now
v3.2.0
September 4, 2026
AI Assistant: summaries that stream, and a transcript you can ask
New
AI Assistant Settings AI Assistant sets the connection once (Ollama, OpenAI-compatible, or Claude) with a Test button that says in one line whether the key and server work. Summaries and dictation clean-up are now separate tasks, each with its own switch, presets, and prompt page, so tuning your meeting-notes prompt never changes what dictation does to a sentence
Ask about this transcript a box under the summary answers questions from the transcript itself ("What was decided about the budget?"), streams the answer, and keeps the thread in the project so it is still there next month
Prompt presets and placeholders start from Summary, Meeting notes, TL;DR, Translate, or Rewrite as article, and use {transcript}, {language}, and {speakers} in your own prompt. Dictation has Punctuation only, Formal, and Casual, with a Try it box to see what a sentence becomes before trusting it with real words
Improved
Summaries stream in the text appears as the model writes it, and the Summary tab opens itself when a summary starts
Long recordings a transcript too long for the model's context is summarized in parts and then as a whole, with a part counter, instead of being cut off
Dictation clean-up runs its own prompt, and if the model fails the raw words are still typed or copied
Fixed
Dictation ran the summary prompt dictated text used to be sent through whatever the summary prompt said, which turned a sentence into meeting notes. Each task now has its own prompt
Settings carry over the old AI settings (key, model, prompt) are migrated on first launch; nothing needs to be entered again
v3.1.11
September 4, 2026
Auto-export: every transcript written where you want it
New
Auto-export turn it on once in Settings Transcription and every transcript is written as soon as it finishes, in the formats you pick (VTT, SRT, text, Markdown, JSON, CSV, DOCX, PDF, HTML), next to each recording, in the projects folder, or in one folder you choose. Existing files are kept unless you ask to replace them, and a recording folder that cannot be written falls back to the projects folder. This brings back what the old batch page did, for the whole app
The batch moment transcribing several files now shows progress across the run with a time estimate, and ends with a card that says how many were exported, skipped, or failed, with an Open folder button, plus one desktop notification for the whole batch. Each file in the list says what it produced
Export menu in a batch the Export button becomes a menu with Export this transcript, Export all finished for anything done before auto-export was on, and the auto-export setting. Thanks to @LivingInternet for the report (#1480)
Improved
Theme switch a sun and a moon in Settings instead of a dropdown
Every audio track is heard a recording with several audio tracks (OBS, Discord, one per speaker) used to be transcribed from one track only; the rest is now mixed in, so every speaker appears and the engine no longer repeats a line over silence it was never meant to hear
Fixed
Repeated sentence for minutes the loop reported on multi-track recordings, where the model heard silence on a dropped track while the speaker kept talking, is gone with the change above (#1518, #1505)
v3.1.10
September 4, 2026
Correct transcripts on AMD graphics with older drivers
Fixed
No Vulkan runtime, no problem a Windows PC without a Vulkan loader (no vulkan-1.dll, typical of machines with no GPU driver installed) could not start the transcription engine at all. The engine now opens the loader at runtime and simply runs on the CPU when there is none. Upstream: ggml-org/ggml#1614
Correct transcripts on AMD graphics with older drivers on some AMD GPUs running a 2020-era Windows driver (seen on a Ryzen laptop's Radeon iGPU), the GPU produced nonsense: a lone "!" or random characters, while the CPU was right. The driver leaves the GPU unclassified in the engine, and the attention kernel then asked for a subgroup width the hardware cannot run. Sona 0.6.5 only asks for it where the driver can grant it, and the GPU transcribes correctly again, several times faster than the CPU. Upstream fix proposed to ggml (ggml-org/ggml#1619)
v3.1.9
September 4, 2026
Transcription is back on older AMD graphics drivers
Improved
Errors now say what the engine printed when Sona dies mid-transcription, the error dialog, the logs, and the bug report carry its last lines instead of a bare "failed to read sona event line", so problems like the one above are identifiable from a report
Fixed
Transcription is back on older AMD graphics drivers since the move to the new engine, machines with a 2020-era AMD driver (Vulkan 1.2, common on Ryzen laptops with Radeon graphics) crashed at the first GPU submit with "vkQueueSubmit: Invalid queue". The driver returns an empty queue handle from vkGetDeviceQueue2; Sona 0.6.4 falls back to vkGetDeviceQueue and the GPU works again. Updating the AMD driver is still a good idea, but no longer required
Older Intel and AMD CPUs are back Vibe no longer refuses to transcribe on a CPU without AVX2. Sona ships its CPU engine twice, an AVX2 build and an AVX baseline build, and picks the one your processor can run when it starts. Any x86-64 machine from 2011 on now transcribes; the "Your CPU is not supported" message is gone
Out of GPU memory no longer kills the run when the graphics card runs out of memory partway through a file, Vibe reloads the model on the CPU, retries that file once, and finishes the rest of the queue on the CPU. That model is remembered as too large for your GPU and loads on the CPU from then on. Thanks to @mikemikimike for the report and the first pass (#1471, #1472)
v3.1.7
What's new?
Name your speakers click a "Speaker 1" label in a finished transcript and type a name. It replaces the label on every line of that speaker, on screen and in every export (text, Markdown, HTML, SRT, VTT, PDF, DOCX; JSON keeps the index and adds speakerName). Names are saved with the project, so they are there when you reopen it
Put a line on the right speaker the same popover moves one line to another speaker, or to a new one. A short line that diarization left with no speaker at all now shows a dashed placeholder so it can be given one. Sona 0.6.2 also attributes such lines to the nearest speaker turn within 1.5 seconds, so far fewer arrive unlabelled in the first place
A Whisper prompt no longer blocks Parakeet or Nemotron those engines take no prompt, and a saved one made every run fail with "does not support text prompts". The prompt is now left out of the request when the loaded model reports it cannot use it (the translate switch too), so it stays saved for when you switch back to Whisper. The Settings prompt field says so when the selected model ignores it
Transcribe several links in one go paste a batch of links into the link box, or add them one by one with the plus button, and they queue up in a list under it. They download in turn; one that fails is skipped and listed at the end, and the rest go straight to transcription. The box clears as soon as the run starts
Links with non-Latin titles no longer fail on Windows yt-dlp can print bytes that are not UTF-8, and reading its output aborted the whole download on the first one. Thanks to @mikemikimike (#1473)
Signed and notarized on macOS and Windows.
v3.1.6
What's new?
Scroll back through a transcript while it is still being written following the live text used to be impossible to escape: scrolling up, or clicking a line to play it, snapped you straight back to the tail on the next segment. The hand-off to the reader keyed off a scroll event that was ignored for 700 ms after any scroll Vibe made itself, and every arriving segment refreshed that window, so during a real transcription it never closed. Following now ends on the gestures you actually make wheel, trackpad, scrollbar, Page Up and picks up again when you scroll back to the bottom
Google Meet detection asks for the permission it needs listed in the 3.1.5 notes, but the change landed just after that tag; this is the first build that has it. Reading a browser's window title requires Screen Recording on macOS, and Vibe checked for it without ever asking. It now requests it once and links straight to the settings pane if you decline. Zoom and Teams never needed it
Signed and notarized on macOS and Windows.
v3.1.5
What's new?
PDF export is a real PDF it used to drive the system print dialog, which on macOS returned before the job spooled and handed the printer a blank page. Vibe now generates the file directly: selectable text, embedded fonts, page numbers, and pages that break on their own
Hebrew and Arabic exports read the right way round the previous PDF engine had no bidirectional text support at all: it resolved one direction per line and reversed every glyph in it, English words and numbers included, so a Hebrew transcript came out backwards. Every export is now checked against the Unicode algorithm before it ships
The export dialog remembers you format, content, theme and both switches persist instead of resetting each time you open it, and the whole panel was rebuilt around one consistent row: nine one-line formats, a light/dark choice for the file itself, and a one-off text direction override
Word documents are set for paper a real font that covers Hebrew and Arabic, a proper type scale, speaker labels that stay with their line across a page break, and no dark page. Their timestamps read forwards again
Transcribe on the CPU a new setting in Settings Models for machines whose GPU driver crashes. The engine always supported it; nothing exposed it
Meetings are detected in about a second the detector waited out a five-second debounce on every change, so a call that had already started stayed invisible for most of a sentence. Quitting a call clears the prompt promptly now too, and the window and process scans it relies on got cheaper rather than more frequent
Google Meet detection asks for the permission it needs reading a browser's window title requires Screen Recording on macOS, and Vibe checked for it without ever asking. It now requests it once and links straight to the settings pane if you decline. Zoom and Teams never needed it
The transcription engine draws the same samples whisper.cpp does sona v0.6.1, which also settles the output: the same audio now transcribes identically twice, where before it drifted. Metal on macOS, Vulkan on Linux and Windows
A long list of recordings scrolls smoothly the sidebar only renders the rows you can see, so hundreds of projects cost what ten used to
Delete every project at once Settings Transcription, under the projects folder
Translations filled in across all 21 languages, including the transcription language lists in Turkish and Swedish, which were almost entirely English
v3.1.4 Beta
What's new?
Vibe notices meetings without interrupting them opt in under Recording settings and Vibe can detect active Google Meet, Zoom, and Microsoft Teams calls from real microphone use. The compact floating prompt stays above the call, does not steal focus, and disappears automatically
Record first, transcribe when you choose recordings are saved immediately as durable projects. Automatic transcription is now optional and off by default; audio-only projects have a polished language-and-options view with a clear Transcribe action
Recording works from anywhere configure a global start/stop shortcut, choose microphone and system audio from the meeting prompt, and manage macOS recording permissions directly in Settings
Long transcripts stay smooth transcript rows and export previews are virtualized, the scrollbar is easier to find, and Jump to playing appears only while audio is actually playing
Playback speed is one click cycle through the useful speeds directly from the player instead of opening another menu
Export is visual and predictable one export flow provides live previews for PDF, DOCX, HTML, Markdown, JSON, and text formats, respects dark mode and RTL, uses sensible filenames and Downloads defaults, and reveals saved files reliably
Project titles behave like text click the title to rename it in place, with the caret exactly where you clicked and the sidebar updating while you type. Generated names now clearly identify Record, URL, and File sources without leaking media extensions
Summaries are easier to control optionally summarize new transcripts, retry or improve existing summaries, use a roomier prompt editor, and send up to 64K tokens of meeting context
Cleaner agent setup the API & Agents page focuses on installing the Vibe skill for Codex or Claude, with a short example prompt instead of cURL and starter-prompt clutter
Autostart stays out of the way launching with the system now opens Vibe hidden, including users who already had autostart enabled
Desktop translations are complete all 22 languages now have the same 574 translated keys, including the new recording, meeting, export, summary, and agent experiences
v3.1.2
What's new?
Your settings stick again every setting reverted to its default on restart in 3.1.1. The file on disk was always correct; the app simply never read it back. If you gave up re-setting your preferences, they are safe now
Failures say why a sidecar that dies now reports its exit code and signal instead of a bare "sona process died", and a crash mid-transcription reads as a crash rather than "error decoding response body". Whisper's own diagnostics reach the log again, after being silently discarded by a no-op callback
Broken model downloads are caught files are verified before they are used, incomplete ones are detected on startup and offered a re-download, and an interrupted download can no longer masquerade as an installed model
Half the memory to load a model measured 294 MB 216 MB on a small model, and roughly one model copy less on large ones. Machines that used to run out of memory mid-load have room now
A CPU without AVX2 is named instead of dying anonymously, Vibe says the processor is unsupported and why
The phone pairing QR scans one wrong character in the QR generator misplaced the alignment patterns on any code past a certain size, and the pairing code was just past it. No scanner could read it
Recordings are 9 dB louder mixing in system audio was quietly halving the level, then halving it again. Every recording made with system audio selected was far quieter than it should have been
Dictation defaults to a two-key shortcut Option+Space on macOS, Ctrl+Space elsewhere, instead of a three-key chord
yt-dlp stops nagging the update prompt appeared on every launch and forgot "Later" each time. It is now a dismissible toast at most once a week, and it offers the update when a download actually fails, which is when it helps
Install the agent skill Settings API & Agents writes it to Claude Code and Codex so it is available in every session, with the transcripts folder and the local API baked in
Windows: the VC++ runtime is version-checked the installer only asked whether any runtime was present, so machines with an old one were skipped and crashed later with no message
The product name and a mistranslated setting fixed across all 22 languages
v3.1.1
What's new?
Record from your phone: scan the QR code in Settings Phone once, and your phone can send recordings to this computer it transcribes them with the model you already have loaded and sends the text straight back
The phone side is a web app you add to your home screen: nothing to install from an app store, no account, and it works on cellular as well as wifi
Audio travels over an encrypted peer-to-peer connection built on iroh phone browsers can't hole-punch, so it passes through a relay that only ever sees ciphertext, never your audio
Phone recordings are saved to Documents/Vibe and show up in Recents like any other transcript, and a recording waits safely on the phone until your computer is reachable again
The phone's language picker comes from whichever model you have loaded, so it always offers exactly what your desktop can actually do
New "Launch at startup" setting, off by default, sitting next to "Keep running in the tray" together they keep the desktop reachable when you reach for your phone
Every new screen is translated into all 22 languages
v3.1.0
What's new?
Redesigned the whole app around one drop-to-transcribe screen, with a recents sidebar you can resize and search
Rebuilt transcript editing: click anywhere on a line and the caret lands where you clicked, Enter/Tab/arrows move between lines, Escape reverts
The line being spoken is lit as it plays, the view follows along, and a timestamp click plays from there
New view options for the transcript text size, timestamps, speaker labels and text direction
AI summaries now run from the main window: automatically after a transcription when enabled, or on demand from the transcript's AI menu, rendered as markdown and saved with the transcript
Re-transcribe from the sidebar now asks for language, model and options first
One picker takes files and folders together on macOS, and folders keep working by drag and drop everywhere
The global dictation shortcut is set by pressing the keys, with ready-made combos for the ones macOS reserves
Optional "keep running in the tray": closing the window leaves Vibe running for global dictation, off by default (thanks @eggie04)
Settings moved out of the browser store into an editable app_config.json agents and people can read and change it, and Vibe picks up the edit immediately
Every number field in settings became a stepper you can hold, type into or reset, with "Never" and "Auto" spelled out
An update waiting is shown on the sidebar toggle instead of hiding in a menu
Fixed uppercase extensions being ignored, so a folder of .MP3 files is no longer empty (thanks @plmelancon)
In-app bug reports now arrive titled with the actual error instead of "App reports bug"
Added Czech (thanks @feretCZ), Bulgarian and German (thanks @nimdassdev); every one of the 22 languages is complete
RTL fixes throughout: the sidebar stays on the window's left, the player timeline drags the right way, and shortcuts read left to right
v3.0.23
What's new?
Fixed Vibe and Sona remaining open after exiting on Windows, which could keep GPU memory occupied and prevent relaunching
Added an Advanced setting to automatically unload inactive transcription models and release RAM/VRAM; set it to 0 to keep models loaded
Protected active uploads, VAD, diarization, and long transcriptions from idle unloadingthe timeout begins only after processing finishes
Improved model reuse so Sona skips reloading an already loaded matching model
Added translations for the model inactivity setting in every supported language
Updated Sona to v0.3.5 with Windows parent-process monitoring and reliable child-process cleanup
v3.0.22
What's new?
Added Nemotron 3.5 and Parakeet TDT v3 model support, including streaming transcription for dictation
Updated Sona to improve transcription and clarified the available transcription options
Migrated the desktop app and website to Paraglide i18n and added translation tests
Improved the model selection flow
Improved the Wall of Love design
Stop Sona cleanly before installing app updates
Made Windows code signing optional and fixed build configuration
Improved website deployment and type-checking reliability
v3.0.21
What's new?
Redesigned the main window and rebuilt Settings with a cleaner sidebar layout
Added toggle mode for Global Dictationpress once to start and again to stop
Added an optional floating dictation indicator so recording status stays visible
Improved Sona startup, model loading, error reporting, and process reliability
Fixed Windows console popups when detecting GPU devices
Improved hotkey transcription errors and saved recording filenames
v3.0.20
What's new?
Sona has been completely rewritten in Rust for improved performance and maintainability
Faster startup and a more streamlined architecture
Built-in speaker diarization, no separate runner required
Simpler build and packaging process for contributors
v3.0.19
What's new?
Remember selected audio device across sessions
Improved CPU compatibility check (clear message if AVX2 is not supported)
Fix batch transcription when folders contain files without extensions
Properly stop background audio process when quitting the app
Various internal improvements and cleanup
v3.0.18
What's new?
Improve transcription failure messages to make issues easier to understand
Update the transcription engine for better stability and compatibility
Fix build issues
v3.0.17
What's new?
Add Stable Timestamps mode for much more accurate subtitle timing in videos, tutorials, movies, and TV shows (can be enabled in More Options)
v3.0.16
What's new?
Fix macOS system audio recording
Improved structured error messages for clearer failures
Internal improvements and dependency updates
v3.0.15
What's new?
Configurable recording save path. choose where your recordings are stored
Improved system audio recording on macOS for better reliability by @Ada-lave in #978
New "Wall of Love" section with international support and better mobile layout
Recent languages are now remembered for faster language switching
Easier copy actions with a new copy button for commands
Visual polish and smoother animations across the app
Better handling of edge cases like missing audio devices
v3.0.14
What's new?
Improved Windows code signing reliability (more robust and consistent signed builds)
Safer and more reliable Windows signing flow using a remote YubiKey signing server
Better Windows tooling setup and verification steps (helps avoid install and update issues)
Cleanup of unused macOS signing configuration
v3.0.13
What's new?
Polished macOS installer design
Windows app is now code-signed for better security and fewer warnings
macOS app is now code-signed for better security and fewer warnings
Better Linux compatibility (updated Ubuntu builds for wider glibc support)
Safer transcription flow now checks audio files before running
Dependency handling improved on Linux for smoother installs
Fixed YouTube downloads on ARM devices
v3.0.12
What's new?
Disable transcribe button when no model is selected with helpful hint
Fix model loading crash for users with non-ASCII usernames on Windows
Graceful fallback to CPU when Vulkan GPU drivers are missing
Better error messages when sona crashes during model loading
Download models link moved to top of settings for easier access
Sona now reports version and commit hash on startup
v3.0.11
Whats new?
Real-time audio visualizer while recording see live feedback as you speak (#950)
Re-summarize option quickly regenerate summaries with one click
Improved dictation dialog more stable global shortcut handling
Fixed Ko-fi dialog not closing properly
Small Linux UI fixes
v3.0.10
What's new?
Add optional GPU device selection
Fix segment timestamps scaling in JSON/CSV
Fix RTL layout issues
Bypass system proxy for localhost connections
v3.0.9
What's new?
Fix diarization (speaker recognition) failing on YouTube downloads and non-16kHz audio files
Fix large file upload failing with "Invalid argument" error
v3.0.8
What's new?
Speaker Diarization
High-quality speaker detection with Parakeet identifies up to 4 speakers automatically.
Global Dictation Hotkey
Hold a shortcut anywhere to record, release to transcribe. Copies to clipboard or types at cursor.
More Format Support
MXF video, M4B audiobook, and CSV export.
Flexible Summarization
Works with any OpenAPI-compatible backend.
Sona Engine Upgrade
File size limit increased from 1GB to 15GB.
v3.0.7
What's new?
Support for much larger audio files (up to 15GB, was 1GB)
More reliable model loading with automatic retry on connection hiccups
Better sona binary detection on Linux (Arch, CachyOS, AUR installs)
Clear error message when no model is selected instead of cryptic crash
Improved error diagnostics for faster issue resolution
Option to disable anonymous analytics in Settings
v3.0.6
What's new?
Major architecture upgrade: Vibe now uses a Sona sidecar instead of an in-app whisper process
More reliable transcription with a local HTTP flow (OpenAI-compatible, streamed)
Live progress + segment updates while transcribing
Better stability and isolation (fewer crashes, easier debugging)
Improved command-line support (better CLI detection + ffmpeg handling)
More stable builds across platforms (Windows + Linux fixes)
You can now see the local API address in Settings
Anonymous analytics + better error tracking (helps catch issues faster)
v3.0.5
What's new?
Add Russian language support for i18n configuration
Enhance model download functionality and set Hebrew-AI model for Hebrew locale
Replace largest Ivrit model with faster Turbo model (almost as accurate as Large v2, better at subtitle segmentation)
Update model conversion instructions for improved setup and efficiency
Smarter transcription! You can now control how the model chooses what to say (sampling strategy + beam size)
Transcribe Folder recursively
v3.0.2
What's new?
What's Changed
update ytdlp and add check for updates by @thewh1teagle in #500
v3.0.1
What's new?
Clean updater files #410
Remove file associations #478
Keep system awake while transcribe/record #337
More logging in setup
Fix pdf export colors #455
Add Vietnamese language
pdf dark mode #455
docs: Add clarity about coreml file usage by @andrewginns in #442
Feat/linux installer by @thewh1teagle in #448
feat: sign tauri plugins by @thewh1teagle in #450
Add zh-HK and fix zh-CN in static locales by @xinbenlv in #460
Add Norwegian Language by @MechanikGamer in #459
New Contributors
@andrewginns made their first contribution in #442
@xinbenlv made their first contribution in #460
@MechanikGamer made their first contribution in #459
v3.0.0
What's new?
Improved privacy policy design and user experience
Added a link to the Vibe website for privacy policy access
Enhanced Linux installation options for a smoother setup
Fixed footer and landing page design improvements
Added installation notes and documentation to help users get started easily
v2.6.5
What's new?
Ollama support for seamless integration!
Improved logging from whisper.cpp with Vulkan.
Updated whisper.cpp for enhanced performance.
Updated Tauri for the latest improvements.
Added support for exporting transcriptions in DOCX format.
Start transcription automatically right after video download.
Fixed crash issue when Documents folder is not found.
v2.6.3
What's new?
Various miscellaneous improvements for enhanced performance.
Bug report updated for better issue tracking.
Fix #333.
Fix #331 by applying default to release profile.
v2.6.2
What's new?
Update whisper.cpp and potentially fix vulkan errors
v2.6.1
What's new?
Update whisper.cpp. should fix some issues with Vulkan
v2.6.0
What's new?
Summarize with Claude API for a more efficient and intuitive experience!
v2.5.6
What's new?
Upgraded to Tauri v2 for improved performance and stability
Enhanced tooltip design for a more user-friendly experience
Improved contrast in dark mode for better readability
Word timestamps are now automatically disabled during speaker recognition
v2.5.5
What's new?
Add option to download audio from popular websites
Fix issue with diarization; ensure word timestamps are disabled if diarization is enabled
v2.5.4
What's new?
Check that Vulkan init works or provide instructions to download older vibe version
v2.5.3
What's new?
Updated Italian translations thanks @rsaleri
Include Vulkan runtime and VC++ Redistributable installer on Windows
v2.5.2
What's new?
Fetch models directly from GitHub, with no more dependence on other cloud services
v2.5.1
What's new?
Improved recording filenames for better organization
Default to storing documentation files in the docs folder
Normalize audio before transcription for improved accuracy
Added support for special languages in speical models by model filename pattern
Enhanced logging for better debugging and issue tracking
Updated Tauri and fixed various package issues
Fix(windows): show whisper.cpp errors in Windows correctly by redirecting stdout/stderr/ experimental
Fix(windows): embed vulkan runtime DLLs
Fix(linux): add vulkan runtime deb packages
Added Vulkan SDK support for older CPUs
v2.5.0
What's new?
Vulkan support for AMD, NVIDIA, and Intel GPUsno need for CUDA, with faster computing
All models available for manual install. see Pre built models
v2.4.0
What's new?
Fix language detection and preference on launch
Fix microphone recording on macOS
Added speaker recognition (diarization)
All models available for manual install. see Pre built models
v2.2.0
What's new?
Added support for Hindi (Thanks @lovishchhabra)
Enabled back OpenCL for Windows, improving performance on compatible hardware
Enhanced GPU device information and set GPU preference in settings for better control
Improved main navigation UI for a smoother user experience
Refined overall UI design for a more polished look
Removed Windows 7 support to focus on more recent versions
Fixed incorrect timestamps by using custom whisper.cpp
Contributors
@lovishchhabra
lovishchhabra
v2.1.0
What's new?
Improved internationalization support with custom locale detection
Added option to transcribe word timestamps
Enhanced macOS DMG installation background
Set GPU preference to high performance on Windows by default
Choose GPU device for improved performance (Thanks @israelxss!)
Enhanced text manipulation with 'replaceAll' feature
Added Windows portable support
Added Swagger documentation for server APIs
Improve Polish translations (Thanks for @GitesHubisz)
v2.0.6
What's new?
Fix linux i18n (Thanks for @oleole39)
Add option to transcribe word timestamps
Add macOS dmg installation background
Set GPU preference to high performance on Windows by default
Max letters per sentence! (Thanks for @sdimantsd)
Contributors
@sdimantsd
@oleole39
sdimantsd and oleole39
v2.0.5
What's new?
Speed up by caching model context instead of reload it everytime (Thanks for @Y-PLONI for suggest it!)
Improve Portuguese translations (Thanks for @josemoura212)
Add Polish translations (Thanks for @GitesHubisz)
Fix typo in Install.md (Thanks for @eltociear)
Contributors
@eltociear
@Y-PLONI
@josemoura212
@GitesHubisz
eltociear, Y-PLONI, and 2 other contributors
v2.0.4
What's new?
Catch whisper.cpp panics instead of crashing
Pretty app version in settings at bottom with nvidia / older cpu labels (Thanks for @josemoura212 for suggestion!)
Fix batch export as json (Thanks for @DArlund for reporting!)
Add deep links. you can add vibe://download?url=<any model url> link to your website for let vibe users download modesl!
Contributors
@DArlund
@josemoura212
DArlund and josemoura212
v2.0.3
What's new?
Add support for m4a (Thanks for @yairl for reporting!)
Fix tabs UI contrast (Thanks for @oleole39 for suggestion!)
Add nvidia, opencl, and rpm support for Linux (Thanks for @thegrasshopper104 for the suggestion!)
Add option to paste model link and download directly in vibe
Contributors
@yairl
@thegrasshopper104
@oleole39
yairl, thegrasshopper104, and oleole39
v2.0.2
What's new?
Add option to record from speakers / microphone! (macOS support included)
Fix audio file encoding issues
Improve French translation (Thanks for @oleole39!)
Add rpm installer
Contributors
@oleole39
oleole39
v2.0.1
What's new?
Add French translation (Thanks for @oleole39!)
Add more Chinese translation (Thanks for @Ifan24!)
Add cli support. use Vibe from console directly!
Add Nvidia v11 and v12 (Cuda versions)
Contributors
@Ifan24
@oleole39
Ifan24 and oleole39
v2.0.0
if CPU is unsupported - Show error message and open URL for fix on Windows
Enable metal framework on macOS for GPU and improve speed by ~40%
Update to whisper.cpp version 1.6.2
Add Nvidia to releases
Add release for older CPUs
v1.0.9
New formats: JSON
Fix saving path
Fix translation in tooltips
Remove focus color from print button
Better toast message on saving files
Fix timestamps for PDF / SRT
v1.0.8
New formats: html, pdf
Show toast message after save file
Option to Print pdf
Add discord button in settings
Option to translate into English (advanced options) | Thanks for hbacelar for the suggestion!
Optimize windows with openblas
More tooltips
Fix extra line in srt
v1.0.7
Fix not a number JS error in realtime preview (Thanks for @Ifan24 for reporting)
Fix when opening settings, keep default transcribe langauge instead of changing it
Imrpve design by making filename in the audioplayer as link when hovering
Show logs in Windows by execute the following in cmd.exe:
set RUST_LOG=vibe=trace
%localappdata%\vibe\vibe.exe
add logs to check if cpu supports f16c instruction
use crate showfile to open files as selected state in file manager
Add Nvidia binary (transcribe 1 hour in less than 5 minuets)
Add non avx binary in case of old CPU crash
v1.0.6
Add Chinese language (Thanks for @Ifan24)
Add batch transcribe (Thanks for @renatoianhez)
Add error boundary in case of fatal render error
Fix Windows file path open crash
Improve language input UI by using option groups
Improve audio player
Add native translation to languages selector
Contributors
@renatoianhez
@Ifan24
renatoianhez and Ifan24
v1.0.5
Add option to drop files to window
Add option to open file by right click -> Open with
file drop animation
Support more audio formats such as opus (Thanks for @renatoianhez for the suggestion)
Add crash log to panic hook
Add option to open app config dir for getting logs
Improve ux by replacing onClick with onMouseDown
Add params info on hover in advanced options (Thanks for @NHLOCAL)
Improve default window size
Improve settings window design
Add Swedish language, (Thanks for @2bbe)
Fix taskbar progress animation
Update to latest tauri core / plugins
Improve offline installation
Option to cancel model download
Contributors
@renatoianhez
@2bbe
@NHLOCAL
renatoianhez, 2bbe, and NHLOCAL
v1.0.4
Smooth window creation
Add min window width and height
v1.0.3
Fixed model customize #54
Allow open settings by click #53
Add option to select whether to play sound / focus window when transcription complete
Use custom eyre with serialize backtrace
Window maximized on first open
v1.0.2
Wider UI
v1.0.1
Fix on windows failed to load model if Username contains Hebrew characters
v1.0.0
GPU optimization for Windows with OpenCL! (x1.5 faster)
Commit hash in system info collection
Fix translation in Hebrew for cancel transcript
v0.0.9
Realtime preview of transcription
Option to abort transcription in progress
Show percentage while transcribing
Show progress when updating
v0.0.8
Add Portuguese language, thanks to josemoura212!
Fix invalid segment error due to invalid utf-8 characters returned from whisper by adding a patch to whisper-rs
v0.0.7
Better errors reports
Upgrade tauri to v2
v0.0.6
Windows shadows
Center window on open
Fix export as txt extension
v0.0.5
Auto updater
Allow export in multiple formats
vibe v0.0.4
Fix: remove model hash verify
Add modal if error happens
Add option to reset app
Improve error reporting to include log
Add MacOS support
Optimize GPU using CoreML
v0.0.3
Settings page
Transcribe parametrs (prompt, temperature, etc...)
Better bug reporting
Option to select / download another models
v0.0.2
Optimize performance by 50%using OpenBLAS
Fix multi language encoding
Add format options as srt vtt and normal
Improve design
Common model path per CLI and Desktop
v0.0.1
remove print