Deep Notes

Deep Notes is a professional, production-ready Obsidian plugin designed to act as a Socratic tutor for your notes. It helps you actively recall information, connect concepts across your vault, and deepen your understanding through recursive questioning. Built for long-term use and ongoing support, Deep Notes transforms note-taking from a collection mechanic into a learning mechanic.

Core Features

Socratic Questioning

• Analyzes your current note and generates:
  1. Knowledge Expansion: Probing questions to test your understanding.
  2. Cross-Topic Connections: Questions linking your note to other relevant notes.
  3. Suggestions: Actionable advice to improve clarity or depth.

"Go Deeper" (Recursive Learning)

• Select any generated question and request a follow-up based on your answer.
• The AI analyzes your response and generates a specific follow-up question.
• Creates a nested thread of dialogue, simulating a real tutoring session.

Active Evaluation

• Scores your answers based on semantic similarity to ideal answers.
• Provides specific feedback and highlights the source text in your note.

Vault Indexing (Semantic Search)

• Builds a local vector index of your vault for semantic search and cross-topic questions.
• Ensures referenced notes exist, preventing broken links.

Multimodal Support (OCR & Vision)

• Scans images in your notes using OCR or Vision Models.
• Parses text from Excalidraw drawings for richer context.

Spaced Repetition Integration

• Schedules reviews based on your evaluation score, supporting long-term retention.

Multi-Mode Study Suite

• Switch between four learning modalities from the sidebar header without leaving the view:
  1. Socratic Depth (default): probing questions with recursive follow-ups.
  2. Active Recall (MCQ): six multiple-choice questions with A–D options and instant exact-match scoring.
  3. Feynman Explainer: challenges you to explain the note's core concept in plain language a non-expert could understand.
  4. Flashcard Deck: generates eight front/back cards with a one-click Markdown export in Anki-compatible ?? format.

LLM-as-Judge Evaluation

• Replaces embedding cosine similarity grading with parallel LLM calls for all answered questions simultaneously.
• Returns a correct, partial, or incorrect rating with a short explanation per question.
• Multiple-choice answers are scored by exact option match, bypassing the LLM call entirely.
• Falls back to cosine similarity scoring when no API key is available.

Hybrid Retrieval Engine

• Runs dense vector search and BM25 keyword search in parallel at query time.
• Merges both result sets with Reciprocal Rank Fusion (RRF) before selecting context.
• Applies Obsidian wikilink graph boosts post-fusion: explicitly linked notes score ×1.20, backlinked notes score ×1.10.
• BM25 index warms from all vault files on plugin load with no API calls.

TurboQuant Vector Compression

• Compresses float32 embeddings to int8 using random sign rotation and Lloyd-Max scalar quantization.
• Reduces in-memory vector storage by 4× with less than 1% cosine similarity loss.
• Search uses integer dot products, approximately 4× faster than float arithmetic on the same hardware.

Incremental Vault Indexing

• Computes a content hash per chunk at index time and skips re-embedding unchanged paragraphs on subsequent saves.
• Removes stale chunks automatically when a note's paragraph count decreases.
• Reduces redundant API embedding calls on large vaults with frequent small edits.

Gemini Context Caching

• Creates a 15-minute server-side context cache for the active note on first question generation.
• Subsequent calls in the same session reference the cache instead of resending the full note, reducing input token cost by up to 90%.
• Falls back transparently to standard calls for notes below Gemini's minimum caching token threshold.

Installation

From Obsidian Community Plugins

1. Open Obsidian Settings > Community Plugins.
2. Turn off "Safe Mode".
3. Click Browse and search for "Deep Notes".
4. Click Install and then Enable.

Manual Installation

1. Download the latest release from GitHub Releases.
2. Copy main.js, manifest.json, and styles.css (if present) to your .obsidian/plugins/deep-notes/ folder.
3. Restart Obsidian and enable the plugin.

Configuration

Go to Settings > Deep Notes to configure your AI provider and other options.

Configuration: Setting Up API Keys

To use Deep Notes with cloud AI providers (Gemini), you must provide your own API keys:

1. Open Obsidian Settings > Deep Notes.
2. Select your preferred AI provider (Gemini or Ollama).
3. For Gemini:
   • Enter your API key in the corresponding field.
   • You can obtain API keys from your provider's developer portal:
     • Google Gemini
4. For Ollama (local), no API key is required, but you must have Ollama running on your machine.

Your API keys are stored securely in your local Obsidian configuration and are never shared.

Using Ollama (Local LLM)

To use Deep Notes with Ollama (local AI models):

1. Download and install Ollama from the official website: https://ollama.com/
2. Start Ollama by running ollama serve in your terminal.
3. In Obsidian, go to Settings > Deep Notes and select "Ollama (Local)" as your provider.
4. Optionally, pull specific models (e.g., ollama pull llama3.2:latest) as needed.

No API key is required for Ollama. All processing happens locally on your machine.

Usage

1. Open any note in Obsidian.
2. Click the Deep Notes icon △ or run the command Deep Notes: Open Deep Notes View.
3. Click Generate Questions to start your learning session.
4. Answer questions, use "Go Deeper" for follow-ups, and evaluate your answers.
5. Schedule reviews and track your progress over time.

Contribution

We welcome contributions from the community! To contribute:

• Fork the repository and create a new branch for your feature or fix.
• Submit a pull request with a clear description of your changes.
• Please follow our code style and add tests where appropriate.

For questions or suggestions, see the contact information in the Support & Authors section.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Support & Authors

If you encounter issues, need help, or want to provide feedback, please open an issue on GitHub Issues or contact the maintainers directly:

• Aaditya Dhungana — [email protected]
• Emmanuel Buhari — [email protected]
• Noah Hong Anh Khoa — [email protected]
• Toby Allan — [email protected]
• Muhammad Zohaib Talat — [email protected]

Buy Us a Coffee

If you enjoy Deep Notes and want to support further development, you can buy us a coffee!