muesli

name	muesli
description	Search and retrieve Granola meeting transcripts using the muesli CLI. Use when the user asks about past meetings, wants to find what was discussed, look up meeting details, or reference meeting transcripts.
user_invocable	false

Muesli - Granola Meeting Transcript CLI

Muesli syncs and searches Granola meeting transcripts locally. It downloads transcripts from the Granola API, stores them as markdown files with a local DuckDB database, and provides search, filtering, and display commands.

Data Location

• Transcripts: ~/.local/share/muesli/transcripts/
• Summaries: ~/.local/share/muesli/summaries/
• Index: ~/.local/share/muesli/index/
• Database: ~/.local/share/muesli/muesli.duckdb

Core Agent Workflow

These four commands are the primary workflow for an agent working with meeting transcripts:

1. Sync — ensure data is up to date

muesli sync                # Download new/updated transcripts
muesli sync --force        # Force re-sync, ignoring cache timestamps
muesli sync --reindex      # Reindex all without re-downloading

Auth tokens are auto-detected from the Granola app on macOS. Run sync before searching if you need the latest meetings.

2. Local — see what meetings exist

muesli local               # List all meetings from local DB (no auth needed)

Output: tab-separated doc_id, date, title (sorted newest first, matches list). Pipe to head for recent meetings or grep for a date range:

muesli local | head -10                   # Last 10 meetings
muesli local | grep "2026-02"             # All February 2026 meetings

3. Find — search for relevant meetings

muesli find "quarterly planning"          # Semantic search (default)
muesli find "quarterly planning" --text   # Full-text search
muesli find "budget" -n 5                 # Limit results (default: 10)

Output: numbered list with title, date, score, and doc ID. Use the doc ID with show.

4. Show — view a meeting's content

muesli show <DOC_ID>                      # Summary, metadata, attendees
muesli show <DOC_ID> --full               # Full transcript

Displays title, date, duration, attendees, labels, and summary (or full transcript with --full). Works offline.

Additional Commands

Search (full-text, returns file paths)

muesli search "quarterly planning"              # Full-text BM25 search
muesli search --semantic "team collaboration"    # Semantic search
muesli search "budget" -n 5                      # Limit results (default: 10)

Output: numbered list with title, date, and file path. Use find instead when you need doc IDs for show.

List (API-based)

muesli list                               # List from API (requires auth)

Output: tab-separated doc_id, date, title. Prefer local for offline/faster access.

Fetch a document

muesli fetch <DOC_ID>                     # Fetch a specific document by ID from API

Query by metadata

muesli query --attendee "Alice"           # Filter by attendee
muesli query --label "Planning"           # Filter by label
muesli query --title "standup"            # Search by title
muesli query --title "standup" -n 10      # Limit results (default: 20)

Stats

muesli stats                              # Meeting statistics from local DB

Summarize a transcript

muesli summarize <DOC_ID>                 # Print summary to stdout
muesli summarize <DOC_ID> --save          # Save to summaries directory

Requires an OpenAI API key (muesli set-api-key).

Configuration

muesli set-api-key                        # Store OpenAI API key in keychain
muesli set-config --show                  # Show current summarization config
muesli set-config --model gpt-4o          # Set summarization model
muesli set-config --prompt-file path.txt  # Set custom summarization prompt

Utility

muesli open                               # Open data directory in Finder
muesli fix-dates                          # Fix file mod dates to match meeting dates
muesli tui                                # Interactive terminal dashboard
muesli mcp                                # Start MCP server for AI integration

Transcript Format

Transcripts are markdown with YAML frontmatter:

---
doc_id: uuid
source: granola
created_at: ISO-8601 timestamp
title: Meeting Title
participants: []
duration_seconds: null
labels: []
generator: muesli 1.0
---

Body contains timestamped speaker turns: **Speaker (HH:MM:SS):** text

Note: Speaker names are often generic ("Speaker") rather than identified by name.

Usage Patterns

Find what was discussed in a meeting

muesli find "budget review" -n 5               # Find related meetings
muesli show <DOC_ID>                            # View the summary
muesli show <DOC_ID> --full                     # View full transcript

Find meetings in a date range

muesli local | grep "2026-02"                  # All February 2026 meetings

Read a transcript file directly

Transcripts are markdown files at ~/.local/share/muesli/transcripts/. Read them directly if needed.

Offline usage

Local commands (local, show, find, search, query, stats) work without API access.

Syncing to another machine with rsync

You can copy the muesli data directory to another machine so it can use all offline commands without Granola API access or a Granola account. The source machine must have already run muesli sync.

Full sync (all features including search and embeddings):

rsync -av --delete \
  ~/.local/share/muesli/ \
  user@remote:~/.local/share/muesli/

Minimal sync (just transcripts, database, and index — skip large model files):

rsync -av --delete \
  --exclude='models/' \
  --exclude='raw/' \
  --exclude='tmp/' \
  ~/.local/share/muesli/ \
  user@remote:~/.local/share/muesli/

What each directory contains:

Directory	Size	Purpose	Needed for
transcripts/	~25 MB	Markdown files	show --full, direct file reads
muesli.duckdb	~6 MB	Document metadata, attendees, labels	local, show, query, stats
index/	~16 MB	Tantivy full-text index + vector store	find, search
raw/	~94 MB	Raw JSON from API	Rebuilding transcripts (optional)
models/	~128 MB	ONNX embedding model	Semantic search (downloaded on first sync)
summaries/	small	Generated summaries	summarize --save output
tmp/	—	Temp files during sync	Not needed

Using a custom data directory on the remote:

# Sync to a non-default location
rsync -av --delete ~/.local/share/muesli/ user@remote:~/muesli-data/

# On the remote, point muesli to it
muesli --data-dir ~/muesli-data local
muesli --data-dir ~/muesli-data show <DOC_ID>

Keeping it up to date (run after each sync):

muesli sync && rsync -av --delete --exclude='models/' --exclude='tmp/' \
  ~/.local/share/muesli/ user@remote:~/.local/share/muesli/

The remote machine needs muesli installed but does not need a Granola account or API token. If you exclude models/, semantic search (muesli find) will automatically fall back to text search.