SuprFlow has two separate jobs: recognize your speech and shape the resulting text. The app uses an on-device speech model. Formatting depends on the provider you select. The installed 1.0.0 (12) screen calls the local default Default, meaning no optional AI formatter is selected. Local (Ollama) can use a model on the same Mac when its endpoint is local. Cloud API sends transcript text and formatting instructions to the chosen hosted provider or custom endpoint.
For confidential material, check the selected mode before you speak. A local speech model does not make a hosted rewrite local. Conversely, a hosted formatter in SuprFlow 1.0.0 does not receive the raw audio file through its formatting request; it receives text.
Start with Default
Open Formatting and select Default. Its visible description is “Essential formatting — punctuation, capitalization, and number cleanup.” Default also performs local cleanup such as filler handling. Test a short synthetic paragraph in the app where you actually write. Check that names, numbers, quotations, and negations survive. If this result is good enough, you may not need another AI model in the loop.
Default means no optional external or Ollama formatter is used; it does not mean the whole app makes no network requests. Release telemetry, account/license checks, model downloads, and updates are separate. The current privacy policy describes these paths; verify them against the signed app build before relying on a local-only workflow.
Use Ollama only after checking the endpoint
The current Ollama Endpoint field starts with http://localhost:11434. That points to a service on the Mac when Ollama is actually running there. The field is editable. A remote IP address, hostname, container bridge, or tunneled endpoint changes where the transcript is sent. Before using confidential text, record the full configured endpoint and model, verify who operates it, and test with synthetic speech. “Ollama” by itself is not a guarantee of on-device processing.
An organization may prefer a centrally operated model rather than a laptop model. That can be a valid policy choice, but it is a network transfer and needs the same recipient, logging, retention, access, and training review as any other service. Do not label it local simply because it uses open model software.
Use Cloud API with a provider-specific decision
The Cloud API screen lets a user choose a provider, model, and API key, or a custom OpenAI-compatible endpoint. The formatter builds a request containing the cleaned transcript and formatting instructions and sends it directly to that endpoint. Provider-specific terms govern what happens there. Check the provider and plan's retention, abuse-monitoring, training, subprocessor, and region terms; a personally purchased account may differ from an enterprise agreement.
SuprFlow 1.0.0 stores provider API keys in app preferences rather than the macOS Keychain, while the SuprFlow app token uses Keychain. A firm should review local secret storage and managed-device policy before asking staff to paste organization keys into this screen. Do not assume the provider keys are Keychain-stored.
Compare three settings with one test
| Setting | Content-processing path in SuprFlow 1.0.0 | What to verify |
|---|---|---|
| Default | Speech recognition and cleanup on the Mac. | Model loaded, output accuracy, local history, telemetry and license traffic. |
Local (Ollama) at localhost | Recognition and cleanup on the Mac; transcript goes to a local Ollama service. | Endpoint truly resolves to the same Mac, model, process logs, and retention. |
| Cloud API | Recognition on Mac; transcript text and prompt go to selected provider or custom endpoint. | Provider agreement, selected model, key storage, endpoint, retention, and training controls. |
Use a fictional 60-word note with a name, number, negation, and bullet list. Record the SuprFlow version, formatter setting, endpoint, and resulting edits for each mode. Do not compare outputs without noting which models ran. A formatting model can make a note look polished while changing the substance; a person must review any professional document before sending or filing it.
Questions people ask
Does “Default” turn off all AI?
It uses the local cleanup path without an optional Cloud API or Ollama formatter. Speech recognition still uses a downloaded machine-learning model on the Mac, and local cleanup still runs. Other app network behavior is separate.
Does Cloud API upload audio?
The formatting request sends transcript text and instructions, not the recorded audio. Speech recognition is on-device in that path. Verify the signed release and any future feature before generalizing the claim.
Can SuprFlow promise my text is never used for training?
Only for a clearly scoped, verified local content path can you say no provider receives the words through formatting. If Cloud API or a remote Ollama endpoint is selected, the recipient's terms and configuration matter. A firm should obtain any needed contractual no-training commitment from the actual recipient.
Related guides
- Can a team enforce local-only formatting?
- Where does my dictation go?
- Does SuprFlow use dictation for AI training?
- Set up SuprFlow on Mac
- Is Ollama formatting really local?
- Offline dictation for accountants and auditors