AI cleanup
Applies to WhisprDesk 1.7.3 · Updated 2026-10-25
AI cleanup is an optional step that tidies your transcribed text, fixing punctuation and small errors while keeping your tone, before it is pasted. It is off by default. When it is on, your transcribed text goes to the AI model you choose: a cloud provider using your key, or a local model on your Mac.
Before you start
Open the WhisprDesk app: click its icon in the Dock, or open it from your Applications folder. The WhisprDesk window opens with Home / History selected in the sidebar. If WhisprDesk is running but you don't see its window, click the WhisprDesk icon in the menu bar at the top right of your screen and open the window from its menu. Settings is at the bottom of the sidebar on the left.
AI text cleanup models are set up in Settings > Intelligence. Transcription models (Whisper, NVIDIA Parakeet, and cloud providers) are set up separately in Settings > Transcription. The two are independent: you can transcribe locally and clean up with a cloud model, or the other way around.
Turn it on
Where: WhisprDesk > Settings > Intelligence > AI Text Enhancement
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click Settings at the bottom of the sidebar.
- Click Intelligence.
- Under AI Text Enhancement, turn on Enable text cleanup.
- A model picker appears below it. Choose Cloud or Local.
- For cloud: pick a provider, paste your API key, then click a model.
- For local: pick a model and click Download, then click the model to select it.
The model picker only appears while text cleanup is on. To turn cleanup off again, go back to Settings > Intelligence and switch Enable text cleanup off.
Cloud providers
| Provider | Examples of models offered |
|---|---|
| OpenAI | GPT-5 family, GPT-4.1 family |
| Anthropic | Claude Sonnet, Haiku, and Opus |
| Gemini models | |
| Groq | Several open models |
| Custom | Your own endpoint. The model list is loaded from it. |
The exact model list is shown in the app.
Local models
Local cleanup models run on your Mac with no internet needed after download. They come from the Qwen, Mistral, Llama, GPT-OSS, and Gemma families and are downloaded when you choose them in Settings > Intelligence. They are stored on your Mac.
What is sent
- Cloud model: your transcribed text is sent to that provider. Your audio is not sent for cleanup.
- Local model: nothing leaves your Mac for cleanup.
Cleanup is separate from transcription. For example, local transcription with cloud cleanup keeps your audio on your Mac but sends the text to your cleanup provider.
Talking to the assistant during dictation
Where: any app, while dictating (needs text cleanup on in WhisprDesk > Settings > Intelligence)
While cleanup is on, you can give an instruction instead of dictating. Start with Hey followed by your agent name, then say what you want, for example "Hey WhisprDesk, make this sound more formal." The AI follows the instruction and pastes the result. Without the agent name, it only cleans up your words.
The default agent name is WhisprDesk.
Where: WhisprDesk > Settings > Intelligence > Voice Agent
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click Settings at the bottom of the sidebar, then click Intelligence.
- Find Voice Agent, type a new name, and click Save. Saving a name also adds it to your Dictionary.
Prompt Studio
Where: WhisprDesk > Settings > Intelligence > Prompt Studio
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click Settings at the bottom of the sidebar, then click Intelligence.
- Scroll to Prompt Studio. It lets you view, customize, and test the instructions used for cleanup. Testing a prompt uses your cleanup model in the same way as a dictation.
Where else this model is used
The cleanup model is also used for:
- AI Actions in Notes (sidebar), such as Generate Notes.
- Note titles created after a file is transcribed with Upload (sidebar).
- Re-transcribe in Home / History (sidebar), if cleanup is on.
To use AI Actions you need a cleanup model selected, which means turning on text cleanup. Note that this also applies cleanup to your regular dictation.
Agent Mode is separate
Agent Mode is a voice chat window with its own model and hotkey, set up in Settings > Agent Mode. It answers you in a chat window instead of typing into your app. It is off by default. See Settings reference.