Open SuprFlow → Speech and read the exact state before retrying. The installed 1.0.0 (12) app labels this screen Models, with Speech recognition model cards and a Storage section below. Downloading model… with a percentage means files are still arriving. Loading model… means the app is preparing a model after download or from its local cache. A failed state needs its actual error message. These states lead to different fixes, and repeatedly clicking models while one is active will not speed up the first setup.
SuprFlow uses WhisperKit for on-device speech recognition. It stores downloaded model files under the user's ~/.suprflow directory. On startup it checks for the selected model there, tries to load a cached copy, and if that load fails attempts a new download. On a first install, network access is needed for the speech model; local recognition can run after the model is ready. WhisperKit's upstream project also distinguishes downloaded model assets from the runtime that loads them.
Diagnose the state before changing anything
| What you see | What it may mean | Useful next check |
|---|---|---|
| Downloading model… and the percentage changes | The selected model is being transferred. | Keep the Mac awake and connected; let this phase finish. |
| Downloading model… stays at one percentage | A stalled connection, blocked model host, or download failure is possible. | Confirm ordinary network access, any VPN or corporate proxy policy, and free storage. Record the percentage and duration. |
| Loading model… after download | The files arrived; the Mac is preparing the model. | Give this phase time. Check that the Mac has free space and is not under heavy memory pressure. |
| No models downloaded yet | The app does not currently list a downloaded model. | Select an available model card while online and watch the progress/error. |
| An explicit failure message | The app caught a load or download error. | Save the exact error text, app version, selected model, macOS version, and network type for support. Do not include dictated client material. |
SuprFlow's Speech screen lists downloaded models under Storage and labels the loaded one Active. A downloaded model and the active model are different. Use the cards actually shown in your installed version. An English-only card is marked EN; if your selected language is not English, choose a multilingual model and test it with harmless speech.
A safe recovery sequence
- Confirm the Mac is supported. Check chip and macOS under Apple menu → About This Mac, then compare with the current SuprFlow requirements. An unsupported device will not be repaired by deleting a download.
- Check available disk space. Open System Settings → General → Storage; Apple documents that screen. Use the requirements estimate conservatively and leave room for the app, model, and macOS working files. Record the actual model shown rather than assuming every model has one size.
- Check the network path. First setup requires a model download. On a managed Mac, a proxy, content filter, or firewall may block the model host. Ask IT to review the signed app and its required download endpoints; do not bypass a firm network policy with a personal hotspot or unapproved installer.
- Allow the current phase to finish. The model screen offers Cancel during download or loading. Cancel only if you intend to stop the current attempt; then use the displayed model card to start a fresh attempt once storage and network conditions are corrected. The app disables other model choices during an active download.
- Quit and reopen after a failed attempt. This can distinguish a stalled UI from a repeatable error. The app may try a cached model and redownload if cache loading fails. If the same error returns, collect it for support instead of repeatedly clearing caches.
- Test with a harmless sentence once active. Click a blank local note and dictate “This is a test of the speech model.” If the model is active but no words appear, check microphone permission and input device in the insertion troubleshooting guide.
Avoid manually deleting ~/.suprflow as a first troubleshooting step. It contains downloaded model assets and may contain other app-managed files. Removing it can force a new download without addressing the original connection, permission, or loading error. Speech → Storage has model-management controls; verify the exact effect in your signed version before clearing anything.
Questions people ask
Does “download complete” mean I can dictate immediately?
Not necessarily. The app downloads a model and then loads it. Wait until the app marks the model ready or active, then make a short test dictation.
Why did a previously downloaded model trigger another download?
At startup the app tries to load a cached model. If that load fails, it attempts to download the selected model again. An app or OS update, a changed model selection, or damaged/missing cache are possible explanations, but the error and installed version must be checked before assigning a cause.
Can I finish first setup offline?
Only if the required model is already present and loads correctly. A new model requires network access for download. Account/license checks and optional hosted formatting have separate network requirements; see the data-flow guide.
Related guides
- Compare the available speech models
- Plan first-download disk space and bandwidth
- Set up SuprFlow on Mac
- Fix recording that does not paste text
- Check local versus cloud formatting
- Where does SuprFlow send dictation?