Daisy

Your first recording

After install, Daisy lives in two places: a small floating widget on your screen (the petal), and a main window you open from the menu bar icon or with ⌘⇧H. This page walks you from the empty Daisy state to a finished, summarised transcript sitting in the folder of your choice.

1. Pick your Library folder

Open Settings → General → Storage and choose the folder where Daisy will save your sessions. Common picks:

  • Your Obsidian vault — Daisy writes plain Markdown with YAML frontmatter, so transcripts become first- class notes you can tag, link, and search alongside the rest of your vault.
  • iCloud Drive — if you want sessions accessible across Macs without setting up Obsidian.
  • A dedicated folder under ~/Documents — clean start, easy to move later.

Daisy will create a Daisy/Sessions subfolder inside whatever you pick, and write one Markdown file per session.

2. Optionally connect your AI provider

Open Connections and pick an AI provider for summaries:

  • Apple Intelligence (macOS 26) — runs fully on-device, no API key, no cloud. Fastest privacy-first option.
  • Anthropic Claude — paste your sk-ant-... key. Highest summary quality for long transcripts.
  • OpenAI — paste your sk-... key.
  • Local MCP— point at any MCP-compatible model you’re already running locally (Ollama, LM Studio).

Keys live in your macOS Keychain. Daisy never sees them — each summary request goes from your Mac straight to the provider.

Skip this step if you only want transcripts, no summaries. You can configure it later.

3. Record

Daisy supports three recording modes:

  • Meeting (orange centre) — captures microphone and system audio together. Use for Zoom, Meet, Telegram, anything that plays audio on your Mac. Tap the widget or hit ⌘⇧R.
  • Voice notes (coral centre) — quick one-off thoughts, mic only. Default hotkey configurable in Settings → Hotkeys.
  • Dictation (lilac centre) — hold a hotkey to talk, release to paste transcribed text at your cursor. Wispr Flow-style.

For your first session, the meeting mode is the most-used path — flip on whatever you want to capture, click the widget, speak for a minute or three. The widget centre turns orange and the petals dance with your voice spectrum.

4. Stop & review

Click the widget again, or right-click → Stop & save. The widget centre fades to a pulsing amber while Daisy:

  1. Transcribes the audio with Whisper Large v3 on the Neural Engine (real-time-ish on Apple Silicon — a 30-minute meeting takes roughly 90 seconds).
  2. Runs speaker diarisation on-device with Pyannote so remote voices are labelled Speaker A, Speaker B, and so on.
  3. If you connected an AI provider in step 2, sends the transcript for summarisation. Otherwise skips this step.
  4. Writes the finished Markdown to your Library folder and opens the session in the main window.

5. What lands on disk

One Markdown file per session, in your Library folder, with a filename like:

2026-05-21 Acme · discovery call.md

Inside: YAML frontmatter (date, duration, tag, attendees if from a calendar event), then the AI-generated summary and action items, then the full transcript with speaker labels. It’s plain Markdown — open it in Obsidian, VS Code, TextEdit, anything.

Where to go next

  • Read Three recording modes for the differences between meeting / voice note / dictation.
  • Set up the local MCP server if you want Claude Desktop or Cursor to read your transcripts.
  • Browse other settings in Settings → General— notifications, audio retention, display name (replaces “Me” in transcripts), tag autocomplete.