Release Highlights

What's New in Community 0.4.0

Windows users with a GPU now get clean, accelerated transcription, summaries can be generated in your own language, and new Qwen3 models keep summarization fully private and on-device. Analytics are now off by default and fully opt-in. If you are new here, start with the docs. This update represents 17 merged pull requests and 48 commits since 0.3.0.

---

Major New Features

Windows GPU Acceleration

Windows users with a GPU now get clean, accelerated transcription. Meetily runs transcription on your Windows GPU through Vulkan, and 0.4.0 keeps Vulkan builds on the correct GPU path on NVIDIA machines instead of mis-enabling FlashAttention. The result is clean, accurate transcripts on Windows GPUs without falling back to CPU. GPU acceleration stays active; only the unstable code path is bypassed.

Multi-Language Summaries

Generate meeting summaries in the language you work in, not just English. Summaries now respect your chosen language and persist that preference across sessions. Transcription already supports 99+ languages and runs 100% locally; this brings the same flexibility to the summary step.

Qwen3 Local Summary Models

Meetily Community now supports Qwen3 models for generating meeting summaries locally. Qwen3 runs fully on your device, so your meetings stay 100% private - nothing is sent to the cloud for summarization. It is another strong open-weight option alongside the existing local AI choices.

Single-Instance App

Meetily now runs as a single instance. Launching the app again focuses the existing window instead of opening a duplicate, so you never end up with two copies fighting over the same recording.

Tray: Hide on Close

Closing the window now hides Meetily to the system tray instead of quitting, so an in-progress recording keeps running. On macOS, reopening from the Dock restores the window.

More Accurate Parakeet Timestamps

The Parakeet engine now uses TDT duration decoding for timestamps, producing more accurate per-segment timing in your transcripts.

---

Quality Improvements & Fixes

• Model Manager responsiveness - the Model Manager UI no longer stalls while models load.
• Summary language normalization - summary language is normalized with a reliable English base, fixing cases where the wrong language could be selected.
• Summary editor render guard - resolved a rendering issue in the BlockNote summary editor.
• Language persistence - follow-up fixes so your transcription and summary language choices persist correctly.
• Faster transcript recovery - shorter grace period before transcript recovery kicks in, so recovered notes appear sooner.

---

Privacy: Analytics Off by Default

Analytics are now disabled by default and fully opt-in. Meetily collects nothing unless you explicitly turn it on. Transcription continues to run 100% locally on your device, and your recordings never leave it.

---

Upgrade & Notes

• Your existing meetings, recordings, and settings are preserved. No migration is required for the desktop app.
• Legacy backend archived: the older Python/FastAPI backend, its Docker setup, and the standalone whisper-server are now archived and unsupported. The supported and recommended way to run Meetily is the desktop app for Windows and macOS. If you were self-hosting the legacy backend, note that it will no longer receive updates. The open source code remains available in the repository for reference.

---

Download

• macOS (Apple Silicon): Download .dmg(opens in new tab)
• Windows: Download .exe(opens in new tab)

New Contributors

• @lorenzojb - Parakeet TDT timestamp decoding
• @passabilities - Tray hide-on-close behavior
• @p-s-vishnu - Multi-language summaries

Full Changelog

View on GitHub(opens in new tab)