It can be, but the endpoint decides. In SuprFlow 1.0.0, Ollama formatting defaults to http://localhost:11434. When Ollama runs on the same Mac at that address, the transcription text goes to the local Ollama server for a rewrite. SuprFlow also lets you change the endpoint. If you enter a company server, another laptop, a proxy, or a tunnel, the transcript travels to that host. The label “Local (Ollama)” does not override the address you configured.
Ollama's own documentation says its server binds to 127.0.0.1:11434 by default and can be exposed on a network by changing OLLAMA_HOST. It also distinguishes locally run models from Ollama's cloud features. Check both the SuprFlow endpoint and the Ollama model actually chosen before using this path for confidential work.
What SuprFlow sends
SuprFlow sends an HTTP POST to the configured endpoint's /api/chat path. Its JSON body includes the chosen model, a formatting instruction, the transcription between delimiters, and generation settings. The normal formatting request sends text, not the recorded audio file. On a local endpoint, that HTTP request stays on the machine's loopback interface; on a remote endpoint, it is a network content transfer. If an endpoint uses plain http:// on a network, the organization must assess transport protection as well.
The speech-recognition step remains separate: SuprFlow uses a downloaded local model to turn audio into text before this formatter runs. Selecting Ollama changes the text-formatting path, not the speech model. If you only need punctuation and simple cleanup, the installed app's Default formatter performs a local cleanup pass without Ollama.
Verify the exact path
- In SuprFlow Formatting, record the provider, the full Ollama endpoint, and the model name.
localhostor127.0.0.1points to the local machine in a normal configuration. A private IP, DNS name, or public URL needs separate review. - Check the Ollama process on that Mac and which model is loaded. Ollama documents
ollama psfor this purpose. A model name in SuprFlow's settings does not prove that it is installed or served by the intended host. - Test with a fictional paragraph containing a number, negation, and unfamiliar name. Compare the source transcript with the formatted output. A local model can still make substantive errors.
- If the endpoint is remote, ask who operates it, what it logs, how it is secured, whether it calls a cloud model, and how retention and training are governed. Record the answer as part of the firm's approved configuration.
- After a SuprFlow or Ollama update, repeat the endpoint and model check. Do not assume a previous local setting still applies.
For a strict device-local workflow, also inspect Ollama's own cloud settings. Its current FAQ describes a local-only mode using OLLAMA_NO_CLOUD=1 or disable_ollama_cloud in its server configuration. That is an Ollama control, not a SuprFlow switch. Confirm it is active in the environment you actually use.
Questions people ask
Does an open model make the formatter private?
The model's license and where it runs are different questions. A model served by another computer receives the transcript over a network even if its weights are open. Check the endpoint and operator.
Does Ollama receive my microphone audio?
SuprFlow's /api/chat request contains the recognized text and formatting prompt, not the audio recording. Verify the signed release if this distinction is part of your firm's approval.
Can I use Ollama for client notes without a provider contract?
That depends on your firm's device and data policy. A model running on an approved Mac may avoid a hosted text provider, but local transcription history, clipboard insertion, destination-app sync, account traffic, and telemetry still need review. A remote Ollama server may require a vendor or internal-service assessment.
Related guides
- Choose local or cloud formatting
- Where does my dictation go?
- Does SuprFlow use dictation for AI training?
- What SuprFlow reads through Accessibility