For a first SuprFlow installation, the current requirements page says 4 GB of free disk space and internet for an initial download of about 2.5 GB. Treat the download as an estimate and the free-space figure as a minimum for the app and its on-device speech models. If you are preparing 20 managed Macs, that is an estimated 50 GB of initial transfer across the fleet before retries or app updates—not a guaranteed invoice or an exact model-file size.
This is a planning question, separate from a stuck model download. SuprFlow 1.0.0 downloads a WhisperKit speech model during onboarding and stores model assets under the user's ~/.suprflow directory. It can later load a cached model and also offers model changes. The public requirements page does not break down the 2.5 GB into installer, model, and other traffic, so a team should measure its chosen signed build once before multiplying the estimate.
A rollout worksheet for IT
| Item | Starting point | What to record in the pilot |
|---|---|---|
| Supported Mac | Apple Silicon and macOS 14 or later on the current requirements page | Actual chip, OS, available space before install |
| Free disk | Published minimum: 4 GB | Space before install, after app install, after model is active, and after one update |
| Initial network | Published estimate: about 2.5 GB | Bytes transferred per device, download host, proxy result, retries |
| Chosen language and model | The app offers selectable model cards | Exact model name and whether another model was downloaded |
| Shared office link | Twenty initial downloads at the estimate total about 50 GB | Peak concurrency, link speed, content-filter logs, user wait time |
| Offline use | The app can recognize speech on-device after its model is ready | Disconnect only after activation and test a harmless sentence |
Leave ordinary macOS working room beyond a published minimum. Do not fill a nearly full managed Mac to within a few megabytes of the threshold. If staff are on metered connections, arrange first setup on an approved network. For a proxy-restricted fleet, test the actual model download on the same network policy users will have; an installer that launches successfully does not prove the model host is reachable.
The current release has a model cache and retry path, but that does not establish a supported shared-model preloading method or a guaranteed enterprise cache. Do not promise that one download can simply be copied between employees. If central staging matters, ask the vendor for a supported deployment procedure and verify it with the signed release. Record the version, model, and date because sizes can change.
What “works offline after setup” covers
On-device recognition and the default cleanup can run without sending dictated audio to a speech server once the model is loaded. That does not mean every app function is network-free: first setup, licensing, updates, release telemetry, and optional cloud text formatting are distinct paths. A custom Ollama endpoint can also point to another machine. See where dictation data goes and local versus cloud formatting before approving sensitive material.
Questions people ask
Is the model exactly 2.5 GB?
The website says the initial app and speech-model download is approximately 2.5 GB. It does not publish a per-model byte count. Measure your selected model and release if exact capacity is important.
Will every employee need their own download?
The app stores models under each user's home directory. A supported shared-cache or preloading workflow has not been verified. Budget per device and user until the vendor confirms a deployment method.
Can employees dictate with no internet after the first download?
Test an active local speech model and default cleanup offline. Cloud formatting, remote Ollama, sign-in, licensing, updates, and telemetry may still need network under their respective conditions. Describe the exact workflow rather than calling the whole product offline.
Related guides
- Fix a model download that appears stuck
- Set up SuprFlow on a Mac
- Check a signed Mac release before a team rollout
- Can a team enforce local-only formatting?