Muesli

A fast, offline-first Rust CLI for syncing and searching Granola meeting transcripts

Muesli syncs your Granola meeting transcripts to local markdown files and provides powerful search capabilities including full-text search (BM25) and semantic search (embeddings).

Features

🔄 Sync transcripts - Download and convert to clean markdown with frontmatter
🔍 Full-text search - Fast BM25 search with Tantivy
🧠 Semantic search - Meaning-based search using e5-small-v2 embeddings
📝 AI summaries - Generate structured summaries with OpenAI
🚀 Fast & offline - All search happens locally, no API calls
💾 XDG compliant - Follows XDG Base Directory specification
🔒 Secure - API tokens in keychain (macOS) or environment variables

Installation

From Release Binaries

Download the latest release for your platform:

# macOS (Apple Silicon)
curl -L https://github.com/harperreed/muesli/releases/latest/download/muesli-macos-aarch64 -o muesli
chmod +x muesli
sudo mv muesli /usr/local/bin/

# macOS (Intel)
curl -L https://github.com/harperreed/muesli/releases/latest/download/muesli-macos-x86_64 -o muesli
chmod +x muesli
sudo mv muesli /usr/local/bin/

# Linux
curl -L https://github.com/harperreed/muesli/releases/latest/download/muesli-linux-x86_64 -o muesli
chmod +x muesli
sudo mv muesli /usr/local/bin/

# Windows
# Download muesli-windows-x86_64.exe from releases page

From Git (no crate publish needed)

# Install Rust (if needed)
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# Install directly from GitHub (all features)
cargo install --git https://github.com/harperreed/muesli.git --all-features

# Or install with specific features
cargo install --git https://github.com/harperreed/muesli.git --features index,summaries

From a Local Clone

git clone https://github.com/harperreed/muesli.git
cd muesli

# Install to PATH with all features
cargo install --path . --all-features

# Or build without installing
cargo build --release --all-features
# Binary is at target/release/muesli

Quick Start

# 1. Set your Granola API token
export GRANOLA_TOKEN="your-token-here"
# Or let muesli read from ~/.granola/supabase.json

# 2. Sync your transcripts
muesli sync

# 3. Search (full-text)
muesli search "quarterly planning"

# 4. Search (semantic - meaning-based)
muesli search --semantic "improving team collaboration"

Usage

Sync Transcripts

# Sync all transcripts (updates only changed documents)
muesli sync

# Force rebuild text search index without re-downloading
muesli sync --reindex

Synced files are stored in:

Transcripts: ~/.local/share/muesli/transcripts/ (markdown)
Raw data: ~/.local/share/muesli/raw/ (JSON)
Indexes: ~/.local/share/muesli/index/ (search indexes)

Search

Full-text search (keyword matching with BM25 ranking):

# Basic search
muesli search "machine learning"

# Limit results
muesli search "product roadmap" -n 5

# Multi-word queries
muesli search "Q1 planning meeting"

Semantic search (meaning-based with embeddings):

# Find conceptually similar documents
muesli search --semantic "team productivity improvements"

# Works with questions
muesli search --semantic "how do we handle customer feedback"

# Finds related concepts (not just exact keywords)
muesli search --semantic "innovation strategy" -n 10

List Documents

# List all synced documents
muesli list

Output format: <doc-id> <date> <title>

Fetch Single Document

# Download a specific document by ID
muesli fetch <doc-id>

AI Summaries (Optional)

# Set OpenAI API key (macOS - stores in Keychain)
muesli set-api-key sk-...

# Or use environment variable
export OPENAI_API_KEY="sk-..."

# Generate summary for a document
muesli summarize <doc-id>

Summaries include:

Key topics discussed
Action items
Decisions made
Follow-up items

Configure Summarization

# Show current configuration
muesli set-config --show

# Change the OpenAI model
muesli set-config --model gpt-4o

# Set context window size (in characters)
muesli set-config --context-window 8000

# Use a custom prompt file
muesli set-config --prompt-file /path/to/prompt.txt

MCP Server

Muesli can run as a Model Context Protocol server, allowing AI assistants like Claude to search and access your meeting transcripts.

# Start the MCP server
muesli mcp

Configure in your AI assistant's MCP settings to enable transcript search and retrieval.

Feature Flags

All features are enabled by default. If you need a smaller binary, you can disable features:

Feature	Description
`index`	Full-text search (Tantivy)
`embeddings`	Semantic search (ONNX, e5-small-v2)
`summaries`	AI summaries (OpenAI)
`mcp`	MCP server for AI assistant integration

Building with Specific Features

# Default build (all features, ~21MB)
cargo build --release

# Core only (sync, list, fetch - ~5MB)
cargo build --release --no-default-features

# With only text search (~9MB)
cargo build --release --no-default-features --features index

# With semantic search (includes text search, ~17MB)
cargo build --release --no-default-features --features embeddings

# With summaries (~11MB)
cargo build --release --no-default-features --features summaries

Configuration

Authentication

Muesli looks for your Granola API token in this order:

--token flag
GRANOLA_TOKEN environment variable
~/.granola/supabase.json (Granola desktop app location)

Data Directory

Override the default data directory:

muesli sync --data-dir /custom/path

API Throttling

By default, muesli throttles API requests (500-1000ms between calls) to be respectful to the Granola API.

# Disable throttling (not recommended)
muesli sync --no-throttle

# Custom throttle range (min:max in milliseconds)
muesli sync --throttle-ms 200:400

How It Works

Sync

Fetches document list from Granola API
Checks local cache to determine which documents need updating
Downloads updated documents (metadata + transcript)
Converts to clean markdown with YAML frontmatter
Writes atomically to disk (crash-safe)
Updates search indexes (if features enabled)

Full-Text Search (BM25)

Documents are indexed with Tantivy during sync
Search uses BM25 ranking algorithm (like Elasticsearch)
Searches both title and body fields
Results ranked by relevance

Semantic Search (Embeddings)

Downloads e5-small-v2 model from HuggingFace (~133MB, cached locally)
Generates 384-dimensional embeddings for each document during sync
Stores vectors in binary format (~1.5KB per document)
Search uses cosine similarity for meaning-based matching
Finds related concepts even without keyword matches

Development

Prerequisites

Rust 1.70+ (install via rustup)
Granola API access

Setup

# Clone repository
git clone https://github.com/harperreed/muesli.git
cd muesli

# Run tests
cargo test

# Run tests with all features
cargo test --all-features

# Run integration tests
cargo test --test workflow_integration --features index,embeddings

# Build debug binary
cargo build

# Run with logging
RUST_LOG=debug cargo run -- sync

Project Structure

muesli/
├── src/
│   ├── api.rs           # Granola API client
│   ├── auth.rs          # Token resolution
│   ├── cli.rs           # Command-line interface
│   ├── convert.rs       # Transcript → Markdown
│   ├── error.rs         # Error types
│   ├── lib.rs           # Library exports
│   ├── main.rs          # Binary entry point
│   ├── model.rs         # Data structures
│   ├── storage.rs       # File I/O and paths
│   ├── sync.rs          # Sync orchestration
│   ├── util.rs          # Helpers
│   ├── index/
│   │   └── text.rs      # Tantivy full-text search
│   ├── embeddings/
│   │   ├── downloader.rs # Model download
│   │   ├── engine.rs    # ONNX embedding generation
│   │   └── vector.rs    # Vector store and search
│   └── summary.rs       # OpenAI integration
├── tests/
│   ├── api_integration.rs      # API mocking tests
│   └── workflow_integration.rs # End-to-end tests
└── docs/
    ├── IMPLEMENTATION_STATUS.md
    └── plans/

Running CI Locally

# Format check
cargo fmt --all -- --check

# Clippy
cargo clippy --all-features -- -D warnings

# All tests
cargo test --all-features

Performance

Operation	Speed	Notes
Sync (initial)	~536 docs in 60s	API rate limited
Sync (incremental)	~1s	Only changed docs
Reindex	~538 docs in 1s	From local files
Text search	<50ms	BM25 index scan
Semantic search	~200ms	First query (model load)
Semantic search	<50ms	Subsequent queries
Embedding generation	~100ms/doc	During sync

Binary Size

Build Type	Size	Configuration
Debug	120MB	Default dev build
Release (default)	34MB	`cargo build --release`
Release (optimized)	21MB	LTO + strip (current)

Optimizations applied:

Link Time Optimization (LTO)
Size-optimized (opt-level = "z")
Debug symbols stripped
Panic abort (no unwinding)

Troubleshooting

"No results found" for text search

The text search index needs to be built:

muesli sync --reindex

"No vector store found" for semantic search

Run sync with embeddings feature to generate vectors:

muesli sync

Slow embeddings generation

This is normal on first sync. The e5-small-v2 model (~133MB) is downloaded once and cached. Subsequent syncs only generate embeddings for new documents.

macOS keychain permission denied

Grant Terminal/iTerm2 keychain access in System Preferences → Privacy & Security.

Contributing

Contributions welcome! Please:

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Write tests for your changes
Ensure all tests pass (cargo test --all-features)
Run clippy (cargo clippy --all-features -- -D warnings)
Format code (cargo fmt --all)
Commit with conventional commits
Open a Pull Request

License

MIT License - see LICENSE file for details

Acknowledgments

Built with Tantivy for full-text search
Embeddings powered by e5-small-v2
ONNX Runtime via ort
CLI powered by clap

Related Projects

Granola - AI notepad for meetings
Obsidian - Perfect for organizing synced markdown notes

Built with ❤️ in Rust

Name		Name	Last commit message	Last commit date
Latest commit History 74 Commits
.github/workflows		.github/workflows
docs		docs
src		src
tests		tests
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
Cargo.lock		Cargo.lock
Cargo.toml		Cargo.toml
Makefile		Makefile
README.md		README.md
_typos.toml		_typos.toml

harperreed/muesli

Folders and files

Latest commit

History

Repository files navigation