MacParsleyThe Summaries tab
Settings → Summaries is where you decide who turns audio into a transcript and who turns a transcript into a summary. It has two halves: Transcription at the top and the Summarization Model below it, followed by the automatic features that go with them.
Transcription
| Control | What it does |
|---|---|
| Transcribe new audio automatically | Starts transcribing as soon as audio arrives (a recording, an import, a link). Turn it off to transcribe by hand with the toolbar's Transcription button. |
| Default Transcription Engine | The engine used for new audio. The menu lists only engines that are ready: Apple's built-in engine, any Whisper or Parakeet model you have downloaded, and Plaud, Pocket or Fireflies.ai once they're set up (Plaud's sign-in, or the key). |
| Configure Engines… | Opens the engine window (below). |
| Corrected Spellings | The fixes MacParsley applies to every new transcript, such as a name that's always heard wrongly. See Transcripts and speakers. |
A line under the controls says where transcription will happen: on this Mac, or in the cloud if you've chosen Plaud, Pocket or Fireflies.ai.
Configure Engines
- On this Mac lists every transcription model with its size and a note. Download fetches a model once; Delete removes it again. Apple (built in) needs no download. Speaker separation uses a small separate model that downloads the first time it's needed. A model appears in the engine menu as soon as it's installed.
- Fireflies.ai holds Fireflies' key, the check that its plan can take audio, and the cloud storage it downloads from. See Fireflies.ai.
Choosing Fireflies.ai, Pocket or Plaud as the engine sends your audio off your Mac. MacParsley asks you to confirm the first time. If the service can't take a recording, that meeting is transcribed on your Mac instead and a message says why.
Summarization Model
The menu shows only what is set up, so it never offers a choice that can't work:
- Models on this Mac or your network: Apple On-Device Model, and LM Studio or Ollama once they have a model.
- Cloud providers once they have a key.
- Fireflies.ai, note1.ai, Pocket and Plaud once their key, token or sign-in is set up. These work from the audio, not the transcript.
- The model you have chosen right now is always shown, even if it isn't fully set up, so the menu is never blank.
- Configure Additional Models at the bottom. It isn't a model: it opens the Model Settings window and leaves your choice as it was.
Under the menu, the chosen model's own settings appear: its key, the model name, an address if it needs one, and a test. For Fireflies.ai, note1.ai or Pocket you see that service's form instead, with a one-line reminder that its summary comes in its own layout, so prompts and styles don't apply. A status dot shows whether the model is ready.
When a service is the Summarization Model, one more line appears: Titles, short mind maps and your own re-summaries use a normal model, which you pick there. A service reads audio and can't write a title or a mind map from text.
Model Settings (Configure Additional Models)
A window with three groups on the left and the settings of the one you click on the right:
| Group | Contains |
|---|---|
| On this Mac or your network | Apple, LM Studio, Ollama. |
| Cloud (the transcript is sent to the company) | Anthropic, OpenAI, Gemini, xAI, Moonshot, DeepSeek, Mistral, Groq, OpenRouter, Perplexity and a custom server. |
| Work from the audio (the recording is sent to the company) | Fireflies.ai, note1.ai and Pocket. (Plaud is chosen in the Summarization Model menu once you're signed in.) |
A green dot means the entry is set up. Make Default (on a model's page) makes it the Summarization Model. The window is also reachable from the Model Settings… button at the bottom of the tab.
Automatic features
| Setting | What it does |
|---|---|
| Write a summary automatically after a recording is transcribed | On by default. Off means summaries are written only when you ask. A meeting that arrives with a summary from a service keeps it and doesn't get a second one. |
| Write a concise mind map with each summary | A model turns the summary into a short mind map. The detailed map is always drawn from the summary with no model. |
| Automatically name new meetings | Gives a recording that only has a file name a written title, after the summary exists. A title you typed is never replaced. |
| Use a larger model on this Mac for long recordings | Shown when Apple's model is the Summarization Model. For recordings longer than about 10 minutes it hands the job to LM Studio or Ollama if one is set up. A cloud model is never used without you choosing it. |
| Let models on this Mac reason before answering | Lets "thinking" models think first. Slower, sometimes better. Off by default. |
Recognising the kind of recording
MacParsley tries Apple's on-device model first, so a recording is never sent anywhere just to work out which prompt fits. If Apple's model isn't available:
- If Apple's model isn't available, use the Summarization Model instead (off by default) sends an excerpt of each new recording to the Summarization Model for that decision. For a cloud provider that means off your Mac.
- With it off, MacParsley skips detection and uses your fallback prompt (set in Settings → Prompts).
How this fits with the rest
- Per meeting. The Summarize window and the Import window can choose another model, or a service, for one meeting. See Summaries and mind maps.
- Per folder. A folder can set its own Summarization (Edit Folder), so sensitive folders use a model on your Mac. It can also name Plaud, Pocket, Fireflies.ai or note1.ai once they're set up, which sends the audio of meetings in that folder to the service. The order is: the meeting's own choice, then the folder's, then the Summarization Model here.
- Choosing between them. See Which should I use?.
- Privacy. What leaves your Mac under each choice: AI models and privacy.