Daisy

Local summaries with Ollama

A local model writes the summary on your Mac: the transcript never leaves it, and the model writes Russian and other languages Apple Intelligence does not. Daisy 1.0.8.34 and newer downloads the model for you — no Terminal, no ollama pull.

Install Ollama once

  1. Open Settings → Summary and pick Ollama as the provider.
  2. If Ollama is not on the Mac yet, Daisy shows Install Ollama. It opens ollama.com/download; Daisy installs nothing itself. Ollama is free and open source.
  3. Install Ollama and open it once. Daisy checks for it every few seconds, and the list of models appears without a relaunch.

Download a model

Each model in the list shows its size and whether it fits this Mac. Daisy reads the memory installed and marks one as recommended:

  • 8 GB — Qwen 3.5 4B (about 3.3 GB)
  • 16 GB — Gemma 4 E4B (about 6.6 GB)
  • 32 GB and more — Gemma 4 12B (about 8 GB)

Press Download. The bar shows the progress; Cancel stops it, and Ollama keeps the part already downloaded, so the next Download resumes rather than starting over. Closing the Settings window does not stop a download; quitting Daisy does. When the model lands, Daisy selects it; Test summary writes a short sample so you can see it works in your language.

Macs with less than 8 GB of memory are not offered a local model: it would be slow and may not fit next to the transcription. A cloud provider is the better choice there.

What leaves the Mac

  • The model file, once, from ollama.com — the size shown on the button. This is the only download in this flow, and it happens only when you press the button.
  • Nothing else. Daisy talks to Ollama over loopback (127.0.0.1:11434); the transcript stays on the Mac. The full list of hosts Daisy can reach is on Verify it is local.
  • Models tagged :cloud on ollama.com run on Ollama's servers, not on your Mac. Daisy's list does not include them, and marks one if you type it in by hand.

If something is off

  • The list does not appear after installing. Open the Ollama app once — it starts the local server Daisy looks for. Check thatServer URL in Daisy is http://127.0.0.1:11434.
  • A download fails. Daisy shows Ollama's reason under the model. Pressing Download again resumes; a model is never left half installed — Ollama lists one only after it is complete.
  • Summaries are slow. Pick the model marked for your Mac, or a smaller one. Long meetings take a few minutes on a small model.