Release Date: April 27th, 2026

v0.2.7 is a big one — four new provider integrations, native text selection on both macOS and Windows, undo cancellation, export support for manual transcriptions, and a long list of quality-of-life improvements.

New Features

xAI Transcription Real-time streaming and batch transcription powered by xAI Aurora. Supports full speaker diarization and 25 languages. Connect your xAI API key in Speech Models to get started.

Speechmatics Transcription Real-time and batch transcription via Speechmatics with speaker diarization and support for 50+ languages. Auto-detect is available for batch transcription; real-time requires a language to be selected.

Cerebras AI Enhancement Connect Cerebras for fast AI-powered transcription enhancement. Supports Llama 3.1 8B and Qwen 3 235B models via the official Cerebras SDK.

AWS Bedrock AI Enhancement Use AWS Bedrock foundation models for AI enhancement. Vowen dynamically discovers available models in your region — just provide a bearer token and region.

Undo Cancellation Accidentally cancelled a recording? Hit Undo within 5 seconds to resume right where you left off. Your audio buffer is preserved, so nothing is lost.

Export Manual Transcriptions Export manual transcriptions to TXT, PDF directly from the transcription detail page. Speaker labels and timestamps are preserved in exports.

Language Selection in Modals Choose the transcription language directly inside the Notes and Manual Transcribe modals. The language list updates dynamically based on the selected model.

Start Minimized to Tray (Windows) Launch Vowen straight to the system tray on startup. Access it anytime from the tray icon without it taking up space on the taskbar.

OpenAI Responses API Format Custom API servers using OpenAI's newer "responses" format are now auto-detected and fully supported alongside the standard chat completions format.

Improvements

Native Text Selection (macOS) Selected text is now read using the native Accessibility API instead of clipboard manipulation. Your clipboard is never touched during text selection, eliminating clipboard corruption.

Native Text Selection (Windows) Selected text is now read using the native UI Automation API instead of clipboard manipulation. Supports modern controls including Edge and other TextPattern2-compatible apps.

Reliable Long Text Insertion (macOS) Long transcriptions are now chunked at sentence boundaries before being inserted. A small delay between chunks prevents macOS from delivering keyboard events out of order — no more jumbled paragraphs.

Sleep Prevention During Notes (macOS) Your Mac will no longer go to sleep while a meeting notes session is in progress. The blocker is released when notes stop, and locking the screen won't interrupt your session.

Notes Summary Language Meeting note summaries are now generated in the same language as the meeting content, instead of defaulting to English. Works across segment summaries, final summaries, and full meeting notes.

Vocabulary for More Providers Custom vocabulary terms are now sent to ElevenLabs (keyterms), Mistral (contextBias), and Sarvam (prompt) for better recognition of domain-specific names and jargon.

Transcript Editor Performance Editing long transcripts is now significantly faster. Text areas are memoized and only the changed segment is resized, reducing layout thrashing during editing.

Streamlined Modals The Notes, Transcribe, and Import modals have been redesigned with portal-based dropdowns, provider sublabels, and a cleaner visual layout.

Tone Disconnection Warning When disconnecting a model that's currently used by one or more tone profiles, you'll now see a warning listing the affected tones.

Dismiss AI Error Banner AI enhancement error notifications can now be dismissed for one hour using the "Hide 1hr" button, so repeated errors don't interrupt your workflow.

Local API Servers Without Keys Custom and local API servers no longer require an API key. If you're running a local setup, just providing a base URL is enough.

Fixes

• Fixed speaker labels not getting saved when the transcript was edited at the same time

• Fixed a bug that prevented retrying a failed transcription from working correctly

• Groq transcriptions that produce repetitive hallucinated text are now automatically detected and retried

• Fixed incorrect transcription providers appearing in the Notes and Manual Transcribe modals

• Custom instructions now clearly indicate when AI Enhancement needs to be enabled first