AI providers (Anthropic, Ollama, Apple Intelligence)

Choose between cloud, local, and on-device AI for notes and Ask AI, all configured under Settings, then AI.

Perchnote does not lock you into one AI provider. AI is pluggable: cloud, local, on-device, or none at all. Pick whichever fits your privacy and quality needs under Settings, then AI.

Anthropic (cloud, default). Uses Claude models over the API. Highest quality, and it costs money per use through your Anthropic account.

Add your API key. It is stored in the macOS Keychain, not in plain text.
Choose a model:
- Opus. Highest quality. Best for complex meetings.
- Sonnet. Balanced default. Strong quality, fast. This is the default.
- Haiku. Fastest and cheapest. Good for very long transcripts where speed matters more than polish.

Ollama (local). Runs an open model on your own Mac. Free, fully private, no API key. You need Ollama installed and running. qwen3 is the recommended model. Quality is lower than the top cloud models but nothing leaves your machine.

Apple Intelligence (on-device). Free, private, and built into the OS. Requires macOS 26 or later with Apple Intelligence enabled. No downloads, no key.

About You. Settings, then AI has an About You field that adds personal context to every AI call, so summaries and Ask AI answers know who you are and what you care about. With Anthropic enabled, Perchnote refreshes it weekly from your meeting notes so it stays current without your input. Editing the field yourself turns the auto-refresh off until you clear it again.

Semantic recall. Separately from note generation, Perchnote can use embeddings to power smarter search across your meetings (finding things by meaning, not just keywords). You can back this with Apple or Ollama, or turn it off for keyword-only search. When on, it runs entirely on-device.