Local privacy
Local audio and video only. No account. Speech recognition and the summary stay on this computer, and nothing is uploaded.
Still early. Fine to try; not a finished product.
The installer is not published yet. Check back soon.
What this computer needs
macOS. macOS 13 or later, Apple silicon (M1 or newer). The text model uses this Mac's GPU (Metal). No separate graphics card. 16 GB of unified memory is comfortable; 8 GB opens the app but gets noticeably slow.
Windows. Windows 10 (1903 or later) or Windows 11, 64-bit. This build runs the text model on the CPU and does not include NVIDIA CUDA, so the bar is higher than on Mac: 32 GB of RAM and a recent 8-core CPU are comfortable, and 16 GB is the floor. No separate graphics card. The download is a setup program. The text model downloads the first time you open the app.
Linux. 64-bit Linux. The installer includes the speech models. The text model is downloaded on first launch, and only on macOS and Windows.
The installer includes SenseVoice Small (Chinese, Japanese, Korean, Cantonese) and Whisper Small (other languages). The 4-bit Qwen3.5-4B text model (about 2.74 GB) is not in the installer: it downloads once on first launch. App updates do not download it again unless the model file itself changes. You can switch to a stronger model, or pick a model file or folder you already have.