MacParsleyAI models and privacy
MacParsley uses AI models for three things: summaries (and the prompt choice), titles, and the concise mind map. By default, transcription and speaker recognition run on your Mac. The exceptions are Plaud, Pocket, Fireflies.ai (transcription engine or summarization model) and note1.ai (summarization model), which work from the audio, so they send recordings off your Mac. This page says which model does what, and what can leave your Mac.
Choosing a model
Not sure which to use? See Which should I use?. The settings themselves are described in The Summaries tab.
Settings → Summaries has the Summarization Model: a provider, and the model to use at that provider. Providers come in two groups, plus the services that work from the audio (below). The menu lists only what is set up; its last line, Configure Additional Models, opens the window where every model and service is set up.
On this Mac or your network
| Provider | Notes |
|---|---|
| Apple On-Device Model | Built into macOS with Apple Intelligence. No setup. Smaller context, so very long summaries may use another local model if you allow it. |
| LM Studio | Runs models you download in LM Studio. Start its server; MacParsley lists the models and loads one with enough room for the recording. |
| Ollama | Runs models in Ollama. |
Cloud providers (the transcript is sent to the company)
Anthropic (Claude), OpenAI, Google Gemini, xAI (Grok), Moonshot (Kimi), DeepSeek, Mistral, Groq, OpenRouter, Perplexity, and Custom (OpenAI-compatible) for any server that speaks that protocol. Add the provider's API key in its model settings. Keys are stored in the macOS Keychain on this Mac, never in your library.
The Custom provider can point at a server on your own network; because MacParsley can't know where that is, it counts it as not on this Mac.
Services that work from the audio
These read the recording, not a transcript, so the audio leaves your Mac. They sit in the same menu once you've set them up, and MacParsley asks you to confirm the first time you choose one.
| Service | Can be | Notes |
|---|---|---|
| Fireflies.ai | Transcription engine, summarization model | Needs a paid plan and a cloud storage it downloads the audio from. |
| Transcription engine, summarization model | Needs a key with write access. The audio goes straight to Pocket. | |
| Plaud | Transcription engine, summarization model | Needs both Plaud sign-ins (the connector and Plaud's website). The audio is uploaded into your Plaud account through Plaud's website, which Plaud doesn't officially support. |
| note1.ai | Summarization model | The audio goes straight to note1. Each recording uses a note1 meeting allowance. |
They have no prompts or styles of their own here, so those choices don't apply to their summaries. If one can't take a recording, MacParsley processes that meeting on your Mac instead and says why.
Other AI options
| Setting | Effect |
|---|---|
| Write a summary automatically after a recording is transcribed | Turn off to summarize only by hand. |
| Write a concise mind map with each summary | Turn off to skip the AI-written map. |
| Automatically name new meetings | Gives untitled recordings a written title. Where possible this is done by a model that stays on your Mac. |
| Use a larger model on this Mac for long recordings | Lets Apple's model hand long summaries to a bigger local model. |
| Let models on this Mac reason before answering | Allows "thinking" models to think first. Slower, sometimes better. |
| If Apple's model isn't available, use the Summarization Model instead | For recognizing what kind of recording it is. |
Where the model for a summary comes from
The one you pick in the Summarize window, otherwise the one set on the meeting's folder, otherwise the Summarization Model in Settings → Summaries. This lets you route sensitive folders to a model on your Mac and everything else to a stronger cloud model.
What leaves your Mac
| What | When |
|---|---|
| Audio | Only if you choose Plaud, Pocket, Fireflies.ai or note1.ai (as the transcription engine or the summarization model, or for one meeting). For Fireflies it waits in your own cloud storage for a few minutes, then is deleted; for Plaud, Pocket and note1 it goes straight to them. Each keeps its own copy in your account until you delete it there. Otherwise never: transcription runs on your Mac. |
| Transcript or summary text | Only when the model in use is a cloud provider. The full text of that recording is sent to that provider. |
| Plaud, Pocket and Fireflies requests | Only if you switch them on and sign in. MacParsley asks those services for your recordings. |
| Web links | When you import a link, MacParsley downloads from it. |
| Corrections, speakers, voice prints, the library | Stay on your Mac. |
MacParsley itself has no accounts, no analytics and no tracking.
Sensitive recordings
Some prompts (medical, for example) are marked Sensitive. Before sending such a recording to a cloud model, MacParsley stops and asks, naming the provider. The concise mind map of a sensitive recording is only ever written by a model on your Mac.
If you use the AI apps (MCP) server
The server runs on your Mac and can only be reached from it. But an AI app you connect (Claude, Cursor…) will send whatever it reads from MacParsley to its own provider. Choose a read-only key unless you want the app to change things, and see AI apps for limiting a key to certain folders.
Keeping things on your Mac
- Choose Apple, LM Studio or Ollama as the Summarization Model, and leave the transcription engine on a model on this Mac. Don't choose Plaud, Pocket, Fireflies.ai or note1.ai: they send audio.
- Set a folder's Summarization to a local model for private subjects.
- Don't add cloud API keys.