Search your notes by meaning.
Search happiness → find notes on gratitude, morning routines, and weekend reflections.

Blue Notes

Blue Notes helps you search your notes by meaning, not just by words. It connects ideas, finds patterns you might have missed, and makes your notes feel more alive and connected.

All semantic search happens locally on your computer, keeping your data safe and offline. The optional AI chat can run completely private with Ollama, or use cloud providers (e.g. OpenAI, Anthropic, Google) with your API key.

Features

• 🔍 Semantic Search – Find notes by meaning, not just keywords
• 🔒 Private & Offline – Searching happens on your computer, no data leaves your device
• 🌍 Multi-Language Support – Works with 100+ languages out of the box
• ⚡ Fast Results – ONNX Runtime keeps semantic search fast on your device
• 📦 Minimal Dependencies – Lightweight footprint with minimal external dependencies
• 💬 AI Chat (optional) – Local AI with Ollama or cloud providers (e.g. OpenAI, Anthropic, Google)

Getting Started

Step 1: Installation & Setup

1. Open Obsidian Settings → Community Plugins → Browse
2. Install "Blue Notes" and enable it

The plugin will automatically set itself up in the background. This includes downloading necessary components and indexing your notes which happens offline. First-time setup takes a few minutes. You'll be notified when it's ready.

Step 2: Start Searching

Simply type // followed by your query in any note to start a semantic search.

Example: Typing //happiness research will show notes about "well-being studies", "positive psychology", and "life satisfaction", even if those specific words aren't in your search.

Step 3: Choose Your Model (Optional)

You can customize which AI model to use for semantic search. Different models offer trade-offs between speed, language support, and accuracy.

Open Settings → Blue Notes → Model selection to choose one of the following:

• Multilingual E5 Small (default) - Efficient multilingual model supporting 100+ languages with excellent performance
• all-MiniLM-L6-v2 - Small, fast, but English-only model (choose if performance matters more than accuracy)

How to Use

Command	Description
// your query	Find notes by meaning from within any note
Semantic search	Find notes by meaning across your entire vault
Find related notes	Show notes semantically related to your current file
Process entire vault	Index all notes (run once at setup, or after major changes)
Process current file	Update the index for just the current note

Optional: AI Chat

Want to ask questions about your notes?

1. Click the chat icon in the ribbon (or use Command Palette → "Open AI chat")
2. Choose your AI provider in settings:
   • Ollama (Recommended) - 100% private, runs locally on your machine (requires good hardware)
   • OpenAI, Anthropic, Google - Cloud-based, requires API key and internet
3. Select which notes to include as context
4. Ask questions like:
   • "What are my main ideas about love?"
   • "Summarize everything I've written about postmodern philosophy"
   • "What connections exist between my work notes and personal goals?"

💡 Privacy First: With Ollama, your chat is completely private and offline—just like semantic search. Cloud providers require internet and send selected notes to third-party services.

Privacy: What Happens to Your Notes?

Local Semantic Search (Always Private)

• ✅ Runs 100% on your computer
• ✅ Works offline after initial setup
• ✅ Your notes never leave your device
• ✅ No tracking or data collection

One-time download: When you first use the plugin, it downloads ONNX Runtime binaries (~50-127MB depending on your platform) and a ~113MB AI model from HuggingFace. After that, everything is local.

AI Chat (Optional Feature)

Local AI with Ollama (Privacy-First):

• ✅ Runs 100% on your computer (requires decent hardware)
• ✅ Works offline after initial model download
• ✅ Your notes never leave your device
• ✅ No API keys or third-party services needed
• 📋 Requires: Ollama installed with models like llama3, mistral, or gemma

Cloud AI Providers (e.g. OpenAI, Anthropic, Google):

• ⚠️ Requires internet connection and API key
• ⚠️ Only sends notes you manually select as context
• ⚠️ Uses your own API key (you control the provider)

You're always in control. Choose between completely private local AI (Ollama) or convenient cloud providers. The chat feature is entirely optional and clearly separated from private search.

Where Files Are Stored

Blue Notes stores all files within your vault:

.obsidian/plugins/blue-notes/
├── models/              # AI embedding models
├── cache/               # Search index
├── onnxruntime-node/    # ONNX Runtime binaries (auto-downloaded)
└── data.json            # User settings

All data stays local and can be safely deleted anytime.

FAQ

Question	Answer
Is this really private?	Yes. Semantic search runs 100% locally. AI chat is also 100% private with Ollama, or you can use cloud providers that only send notes you explicitly select.
Do I need an API key?	No. Semantic search and Ollama chat work without any API key. You only need one for cloud chat providers (e.g. OpenAI, Anthropic, Google).
Does it work offline?	Yes. Semantic search and Ollama chat work offline. Cloud chat providers need internet.
Does it work on mobile?	No, desktop-only.
Will it slow down Obsidian?	Only during initial processing. After that, searches are very fast.
What hardware do I need for Ollama?	At least 8GB RAM recommended. 16GB+ for larger models. Modern CPU or GPU helps with response speed.

Troubleshooting

ONNX Runtime installation failed

• Check your internet connection
• Open Settings → Blue Notes → ONNX Runtime Status
• Click "Download ONNX Runtime" to retry the download
• Verify your firewall isn't blocking the download
• Check if you have enough disk space (50-127MB required)

Search returns no results

• Try running "Process entire vault" again
• Try different search terms or lower the threshold in settings

Model won't download

• Check your internet connection
• Download the default model manually
• Download the all-MiniLM-L6-v2 model manually

Plugin feels slow

• Normal for first-time processing
• Faster computer = faster indexing

AI chat not responding

• Check your API key in settings
• Make sure you have internet connection if using a cloud provider
• Verify you've selected notes as context

For Contributors

This project uses changesets for automated version management:

1. Make your code changes
2. Create a changeset: pnpm changeset
3. Commit the changeset file with your changes
4. Create a pull request

For Maintainers

When a PR with changesets is merged to main:

1. GitHub Actions automatically creates a "Version Packages" PR
2. Review and merge the "Version Packages" PR
3. GitHub Actions automatically:
   • Creates a git tag
   • Builds the plugin
   • Creates a GitHub release with assets

Get Help

• Bug reports: GitHub Issues
• Questions: GitHub Discussions

Credits

Built with ONNX Runtime and ❤️ for Obsidian