S1-mini by Superwhisper
S1-mini is a transcript cleanup model. Freehand sends it the completed text from Voice or an audio file, then uses the cleaned result for delivery. Your transcription connection and model stay separate.
What the profile adds
Section titled “What the profile adds”Selecting S1-mini by Superwhisper in Cleanup options for Voice transcription or Audio file replaces the custom instruction editor with the model’s trained controls:
| Control | Choices |
|---|---|
| Styling | Casual, semi-casual, semi-formal, formal |
| Structure | Prose, lists |
| Context | General, email |
Freehand builds the fixed instruction and control line for you. Settings shows the effective prompt for review. Temperature stays at zero, and reasoning must be off. The optional output-token limit and timeout remain available.
Open the workflow’s Settings cog to show Cleanup options on the right. Choose Save to apply your edits; closing or leaving unsaved options offers Save, Discard, or Keep editing. Switching models preserves your cleanup style, structure, and context choices.
Choose a backend
Section titled “Choose a backend”For local cleanup on supported Windows and macOS computers, follow managed llama.cpp setup. Freehand installs the runtime and downloads S1-mini only when you choose those actions. Select its built-in Connection for Cleanup and enable cleanup; the runtime supplies the S1-mini profile and disables reasoning.
| Backend | Reasoning requirement |
|---|---|
| llama.cpp | Freehand sends reasoning_effort: "none"; keep --reasoning off in the launch command and use a template that honors it |
| vLLM | Freehand sends reasoning_effort: "none"; use the qualified version and a compatible template |
| Generic OpenAI-compatible | Configure reasoning off on the server; Freehand shows Disable on server |
For these manual services, select the saved connection in Cleanup options, enable cleanup, choose the model ID exposed by the server, and select S1-mini by Superwhisper as the model profile. Choose your output controls and save.
Language and results
Section titled “Language and results”If cleanup fails, returns empty text, or reports an output-token limit, Freehand uses the raw transcript. Cleanup runs once per completed transcript; it does not split long inputs into chunks. Optional history can retain both the raw and cleaned versions during the current session.