FAQ
Questions before you install.
Short answers.
The hardware, language, Free and Pro and feedback answers match the NexusVox repository. The licence answer quotes the licence text. Where a page here has more detail, the answer links to it.
What stays on my PC?
With the built-in models, your audio and your transcripts stay on your PC. There is no account, and once a model is downloaded NexusVox works offline. The models themselves are downloaded once, on first use.
Audio goes nowhere with the built-in models. If you choose a transcription, language-model or translator server that runs on another computer, what you send to it goes there.
What hardware do I need?
Windows 11 only. NexusVox does not run on macOS or Linux, and it needs Python 3.11 or newer.
With an NVIDIA GPU (recommended): you need the GPU with the CUDA runtime installed, plus Docker Desktop and the NVIDIA Container Toolkit. Parakeet runs in Docker and needs about 2 GB of video memory; Whisper Large V3 Turbo needs about 4 GB. With Parakeet, an 8 s dictation takes 0.08 s on an RTX 5070 Ti; a slower card will be slower.
Without a GPU: no Docker, and Parakeet runs in-process on the CPU. The floor is 4 threads and 4 GB of free RAM, and that is slow; 8 threads and 8 GB are recommended. On the test PC, pinned to 8 threads, an 8 s dictation takes 0.51 s; pinned to 4 threads it takes 2.5 s, and a slower CPU will be slower still. Speaker detection needs the GPU container. To upload audio files other than WAV you also need ffmpeg on your PATH.
Polish and auto cleanup also need a language-model server that you run yourself, Ollama by default. It is not shipped with NexusVox and uses memory on top of the speech model.
The three timings are for Parakeet, each measured on one test PC; the features page lists their sources and test setups. The memory, thread and RAM figures are requirements from the core repo, not measurements.
Which languages work?
Parakeet, the default model, covers 25 European languages and detects the language automatically. Whisper Large V3 Turbo covers 99 languages; switch to it in the dashboard for a language Parakeet does not know.
Whisper Large V3 Turbo is the only Whisper model in the app. It is quick on a GPU. We have no CPU timing for it, so without a GPU expect a longer wait than with Parakeet; it is also a larger download. The timings in the hardware answer are for Parakeet.
What does the licence allow?
NexusVox is source-available under the PolyForm Noncommercial License 1.0.0. This answer quotes the licence and adds no reading of its own.
The licence says: “Any noncommercial purpose is a permitted purpose.” Of personal use it says: “Personal use for research, experiment, and testing for the benefit of public knowledge, personal study, private entertainment, hobby projects, amateur pursuits, or religious observance, without any anticipated commercial application, is use for a permitted purpose.”
Of organisations it says: “Use by any charitable organization, educational institution, public research organization, public safety or health organization, environmental protection organization, or government institution is use for a permitted purpose regardless of the source of funding or obligations resulting from the funding.”
Whether your use falls under these terms is not answered here. For a question about a specific use, write to NightShift AI GmbH at info@nightshift-ai.de. The licence text is the only binding wording, and this is not legal advice. The speech models are not covered by it: each is downloaded under its own terms.
How do Free and Pro differ?
The core stays free and source-available. That includes dictation, polish, auto cleanup, file upload transcription, speaker detection and export as TXT, SRT and VTT. Nothing listed here will become paid.
Pro is a separate, commercial extension that adds to the core and takes nothing away from it. It is available from December. It adds summaries, speaker handling, search and ask, and further export formats.
How do I send feedback?
In newer builds of the app (after v0.7.0), the dashboard has a Feedback tab with a “Send feedback or report a bug” button. It opens the feedback form with your app version, Windows version and hardware filled in, and sends them only if the box stays ticked.
Without the app, use the feedback form here to report a bug or suggest a feature. It reaches the team privately by email. We read every message, but we do not reply to messages sent through the form.
If you want an answer, write to nexusvox@nightshift-ai.de. Bug reports and ideas are also welcome as issues on GitHub.