Two different jobs use two different kinds of model, and it is worth keeping them apart:
- Transcription turns speech into text. It is built in and runs on your Mac. See transcription models.
- A language model writes the summary and answers your questions in chat. That is this page.
Ghosty Notes does not ship a language model
You choose one and connect it. This is deliberate: bundling a model would mean picking one licence, one size and one quality level for everybody, and shipping a couple of gigabytes to people who already run better models locally.
It also means summaries and chat do nothing until you connect something. If your notes are not being summarised, this is almost always why.
The local route, and what we recommend
Run a model on your own Mac with LM Studio or Ollama. Both are free, neither is bundled, and with either one your meeting text never leaves the device.
LM Studio has a normal Mac interface and is the easier start.
- Install it from lmstudio.ai.
- Download a model inside LM Studio.
- Start its local server.
- In Ghosty Notes: Settings, AI, and choose LM Studio.
Ollama is the command line route.
- Install it from ollama.com.
- Pull a model, for example
ollama pull llama3.1. - Ollama serves it automatically.
- In Ghosty Notes: Settings, AI, and choose Ollama.
Ghosty Notes talks to both on your own machine, over localhost. Neither needs an API key and neither sends anything to the internet.
Pick a model with room to read. A meeting transcript is long. A model with a small context window will refuse it. If chat returns a context length error, see the fix.
The API key route
If you would rather use a hosted model, Settings, AI accepts your own key for OpenAI, Anthropic, Google Gemini, Mistral, OpenRouter, Cloudflare Workers AI, Azure OpenAI and Azure AI Foundry. There is also a Custom option for any endpoint with an OpenAI-compatible API.
You are billed by that provider, not by us. The key is yours.
What this sends: the meeting text goes to the provider you chose, so it can write a summary or answer a question. Your audio never does. See privacy in detail.
Which should you use
| You want | Use |
|---|---|
| Nothing to leave your Mac | LM Studio or Ollama |
| The simplest possible setup | LM Studio |
| The best quality, and you accept text leaving the device | Your own API key |
| No summaries at all | Nothing. Recording and transcription work regardless. |
Do I need Pro for this?
No. Bring Your Own Key and on-device AI notes are on the free plan. Pro adds Google and Outlook calendar connections, speaker identification, and human support. See pricing.