Muesli for Obsidian

Granola, in Obsidian. The AI notepad for back-to-back meetings.

Notes, actions and memory, directly in your vault. Without a meeting bot.

English · Français

Install · Setup · How it works · Privacy

Why Muesli?

You jump from call to call. You want to listen, jot three words, and walk out with clean notes and a task list. Not a bot in the participant list, not another app, not your notes locked in someone else's cloud.

• No bot joins your call. Muesli records your microphone and your Mac's audio output, so it works with Zoom, Meet, Teams, Slack, a phone call in the browser — anything that plays sound.
• Your notes come first. Type a few words while people talk. The summary is built around them, filled in with what was actually said.
• Actions you can track. Tasks come out as - [ ] Paul — send the recap (Friday), ready for Tasks, Dataview and Bases.
• Memory that stays yours. Every meeting is a folder of plain Markdown in your vault: searchable, linkable to people and projects, synced however you already sync.
• Bring your own key. One OpenRouter key, pay per use (about $0.10 per hour of speech). No subscription, no account with us.

What you get

Click the 🎙️ ribbon icon, pick Meeting, and write in the Notes.md that opens. Click stop. A few seconds later:

Meetings/2026-10-09 1430 Pricing review with Acme/
  Summary.md      ← TL;DR, decisions, actions, topics, open questions
  Transcript.md   ← cleaned-up "who said what", with timestamps
  Notes.md        ← your raw notes, untouched
  audio.mp3       ← optional recording

The title is chosen by the AI from the content. Speakers are named from the conversation ("thanks Paul…"), and you confirm or fix them in one dialog.

Two uses

During a meeting

• Notes with timestamps. Every line you type is silently timestamped, and Mark this moment drops a ⏱ 12:34 marker. The summary uses both to align your notes with the conversation.
• Never forget to stop. A small always-on-top pill shows the recording, the elapsed time and the estimated cost over every app, and stays out of screen shares and screenshots. After a long silence Muesli asks whether the meeting is still on, and ends it by itself if nobody answers.
• Ready when you stop. Audio is transcribed in the background during the meeting, so the result arrives seconds after you click stop.

Recorded elsewhere? Import it

A voice memo from your phone, an .m4a from a dictaphone, an .mp3 someone sent you: drop it in Meetings/Inbox (AirDrop, iCloud Drive, Finder), right-click any audio file in the vault → Create a meeting from this audio, or run Import an audio recording to pick a file anywhere on disk. You get the same folder, with the original audio kept.

iPhone Voice Memos, processed on your Mac

Turn on Process new memos automatically (macOS): every new Voice Memo synced by iCloud is transcribed, recognized as a meeting or a journal entry from what is said, and filed — while Obsidian is open. Record with the screen locked, find it in your vault. Needs Full Disk Access for Obsidian.

On your phone

Tap the mic icon to open the recorder: pick Journal or Meeting (in person, microphone only), then one big button. iOS cuts the microphone of any app in the background, so keep Obsidian in front while recording (the screen stays on). The OpenRouter key is stored per device: enter it on the phone too.

Install

From Obsidian (recommended): Settings → Community plugins → Browse, search Muesli, then Install and Enable. Or open Muesli in the community directory and select Add to Obsidian.

With BRAT (latest builds): install BRAT, then BRAT: Add a beta plugin → Azilone/obsidian-muesli.

Manually: download main.js, manifest.json and styles.css from the latest release into <your vault>/.obsidian/plugins/muesli/, then enable Muesli.

From source: see CONTRIBUTING.md.

Requirements

Obsidian 1.13 or later, and an OpenRouter API key with a few dollars of credit.

Setup

On first launch, a setup screen walks you through it (also available from Settings → Muesli → Setup or the Set up Muesli command):

1. OpenRouter API key: get one at openrouter.ai/keys and paste your key. It is checked right away, the remaining credit is shown, and it is saved in Obsidian's secret storage (never in your vault or in data.json).
2. Your name: used for your microphone in transcripts.
3. Sound test (Mac): Muesli sets up its small audio helper (see disclosures), then macOS asks to allow Muesli to record the computer's audio. Accept, or the other participants won't be captured. Play some audio during the test: both meters should move. You can change the permission later in System Settings → Privacy & Security → Screen & System Audio Recording. macOS may ask again after an update.
4. Optional: Default meeting type (call or in person), default summary template, summary language, and the folders.

That's it: ribbon icon → Meeting → write → stop.

Summary templates

Templates are plain notes in Meetings/Templates. A template's content describes the summary you want, in your own words. Two are created for you, General and 1:1. Duplicate one, rewrite its sections (a sales call, a standup, an interview…) and pick it as the default in the settings.

Language

The interface, the default folders, the templates and the AI prompts follow Obsidian's language: French if Obsidian is in French, English otherwise. Transcription auto-detects the spoken language (or set it), and the summary is written in the Summary language setting, whatever the meeting language.

How it works

Everything is written to your vault first (<meeting>/.raw/), so a crash or a failed API call never loses a meeting. Re-process the meeting (command, or right-click the folder) picks up where it stopped.

Cost

You pay OpenRouter directly. As a rough guide: transcription is about $0.10 per hour of speech after silence removal, plus a few cents of LLM per meeting. The live estimate is shown in the overlay, the real cost when processing ends.

Settings

Privacy and disclosures

• Network: audio and text are sent only to OpenRouter (and the model providers it routes to), with your own key. No telemetry, no analytics, no server of ours.
• Paid service: OpenRouter is pay-per-use; you need an account and credit.
• Native helper (macOS): system audio capture and fast audio decoding use meeting-tap, a small open-source Swift program (native/macos/main.swift). Obsidian only installs main.js, manifest.json and styles.css, so the release build embeds the compiled helper in main.js (with its SHA-256) and Muesli writes it to its plugin folder on first use. Nothing is downloaded. The helper is compiled from that source by a GitHub Actions workflow and committed as native/meeting-tap, and main.js is built from the repository by another; both carry build provenance attestations, so main.js can be rebuilt and checked byte for byte. You can also build the helper yourself with ./native/build.sh.
• Files outside the vault: only when you import a file from disk, or with the optional Voice Memos inbox (reads ~/Library/Group Containers/group.com.apple.VoiceMemos.shared, requires Full Disk Access).
• Raw audio chunks and API responses stay in a hidden .raw/ folder inside each meeting folder. Delete it once you're happy with the result. Syncing your vault with git? Consider adding **/.raw/ and *.mp3 to .gitignore.
• Always tell participants you are recording, and follow the consent laws where you live.

Development

bun install
bun run check                        # type-check, Obsidian lint rules, tests
./native/build.sh                    # rebuild native/meeting-tap (universal, ad-hoc signed)
VAULT="/path/to/vault" bun run dev   # watch mode, builds straight into the vault

Test the helper on its own:

./native/meeting-tap --seconds 5 > out.pcm
ffplay -f s16le -ar 16000 -ac 1 out.pcm

The helper writes raw PCM (s16le, mono, 16 kHz) on stdout and JSON status lines on stderr, and exits when its stdin closes. meeting-tap --decode file.m4a streams any audio file in the same format (used for imports). Any helper that follows this contract can be added for Windows (WASAPI loopback) or Linux (PipeWire monitor) — contributions welcome.

Layout

src/
  main.ts              plugin, commands, status bar, recovery
  i18n.ts              every UI string, English and French
  session.ts           recording session, transcription queue, on-disk state
  openrouter.ts        API client (STT, chat, key check)
  recorder/track.ts    VAD, silence removal, chunking, time maps
  recorder/sources.ts  microphone (AudioWorklet) and system audio (helper)
  recorder/helper.ts   writes the macOS helper embedded in main.js
  recorder/decode.ts   audio file → 16 kHz PCM (helper `--decode`, Web Audio fallback)
  recorder/voice-memos.ts  iPhone Voice Memos inbox (macOS)
  pipeline/            merge, speaker matching, clean-up, summary, journal, prompts (EN + FR)
  ui/                  start and import dialogs, speaker names, overlay pill, mobile recorder
native/macos/          Swift Core Audio tap helper

Roadmap

• Windows and Linux system audio helpers
• Live transcript while recording
• "Ask your meetings" chat across the vault
• Auto-title and participants from the calendar

License

MIT