Skip to content
← Help

Running Daisy fully offline

Daisy already does the heavy lifting on your own computer: recording, transcription, and speaker diarization all run locally. The one piece that normally reaches the internet is the summary step. This article covers the two ways to keep that local too.

(Aside from the AI step, Daisy's only network touches are a periodic license check, update checks, and activation — never your meetings.)

Pick how summaries are produced

You have several choices for the summary step. The first two keep the summary local too:

  • Local AI server (advanced) — point Daisy at LM Studio or Ollama running on your own machine or network. Setup details below.
  • Skip and use copy-paste — keep transcripts only, paste them into your own ChatGPT or Claude when you want a summary. No setup, no per-meeting cost, no key.
  • Cloud provider with your own key — Anthropic, OpenAI, or Groq. Best summary quality, but the transcript text (not your audio) is sent to the provider over HTTPS with your key.
  • Daisy Cloud — a managed option if you'd rather not run a local model or bring your own key. It's optional, free for now during a limited public release with no warranties, and only the text is proxied — your transcripts and speaker labels stay local, and no conversations are stored. See What is Daisy Cloud?.

Set up a local AI server (advanced)

Daisy can hand summarization off to any OpenAI-compatible local server. The two common options are LM Studio and Ollama.

  1. Run a local server such as Ollama or LM Studio and load a capable model. A 7B–13B parameter instruction-tuned model is the minimum for usable summaries; better hardware means you can run a larger, better model.
  2. In Settings → Providers, configure LM Studio or Ollama under Summarization by entering its Base URL (for example, http://localhost:11434/v1 for Ollama). No API key is needed.
  3. In Settings → Behavior, set Default summary provider to that local server.

Don't want to run a local AI server? Use copy-paste.

If you want full privacy for the transcript but don't want to set up a local AI server and don't want a paid API key, there's a free middle ground:

  1. Open the meeting after Daisy finishes the transcript.
  2. Click Copy all in the transcript view.
  3. Paste it into your own ChatGPT or Claude session, ask for a summary.
  4. Paste the summary back into Daisy.

No setup, no key, no per-meeting cost. See meeting summaries for the paste-back step.

Good to know

  • Your finished transcript and speaker labels are already produced on-device — no cloud account needed. With the network off, live captions run on-device on desktops and Apple Silicon Macs; on a lighter laptop the cloud live-caption option isn't available offline, so keep live captions on-device or just take the full transcript at the end.
  • Local AI servers are keyless. Daisy checks that the server is reachable rather than verifying any credential.
  • Summary quality on a local AI server depends entirely on the model you run. A cloud provider with a frontier model (Claude, GPT-4-class, etc.) will produce noticeably better summaries.