Obsidian-Smart2Brain Project Introduction
Your Smart Second Brain
Obsidian-Smart2Brain is a revolutionary free and open-source plugin designed for the Obsidian note-taking application. It aims to enhance knowledge management by acting as a personal assistant for its users. What makes it particularly innovative is its use of large language models (LLMs) like ChatGPT or Llama2, which enables it to process and interact with notes more intelligently. Unlike many other tools, Smart2Brain can function entirely offline, ensuring the security and privacy of user data.
Features
Chat with Your Notes
- RAG Pipeline: This feature ensures that your notes are embedded into vectors, which are then used to retrieve information based on the similarity of your queries with the content of your notes. This makes answers specific and relevant.
- Reference Links to Notes: As answers are generated, Smart2Brain provides reference links to the origin notes, allowing users to trace back the source of the information.
- Chat with LLM: Users can choose to disable the feature of referencing notes, allowing the plugin to generate answers based on the LLMs' training knowledge.
- Save Chats: Conversations can be saved for later continuation, ensuring ongoing productivity.
- Different Chat Views: Users can select between a 'comfy' or 'compact' chat view, tailoring the experience to personal preferences.
Choose Any Preferred Large Language Model (LLM)
- Ollama Integration: Ollama allows LLMs to run locally, akin to how Docker operates, providing flexibility in use through interactive shell, REST API, or Python library.
- Switch Between LLMs Quickly: Users can switch between different LLMs for various writing purposes, whether scientific or persuasive.
- ChatGPT Compatibility: Even though the focus is on privacy, users can still access the advanced capabilities of OpenAI's models.
Limitations
- Performance Variation: The effectiveness depends on the chosen LLM, as they're designed for different tasks.
- Knowledge Structure Dependency: The quality of responses improves when notes are well-structured and organized.
- Potential for Incorrect Answers: Sometimes, due to insufficient relevant notes or the LLM's limitations, the assistant might provide unsatisfactory answers, which can often be improved by rephrasing the query.
Getting Started
Users should note that using Obsidian Sync might cause an increase in storage usage due to version history. Excluding the .obsidian/plugins/smart-second-brain/vectorstores folder in settings can help manage space efficiently.
Under the Hood
For those interested in more technical aspects, they can explore the project’s Architecture Wiki.
Roadmap
Exciting features are planned for the future, including:
- Support for additional LLMs like Gemini and Claude models.
- New ways to view and connect similar notes.
- Introduction of Chat Threads and an Inline AI Assistant.
- Advanced note prediction and placement technologies.
- Benchmarks for performance evaluation.
About Us
This plugin originated from a university project, driven by a passion for improving AI technology and the Obsidian user experience. It continues to evolve, providing an experimental ground for AI development.
Support and Engagement
Users are encouraged to participate in the project's advancement by reporting issues, proposing features, or contributing to the code via GitHub.
Frequently Asked Questions
Queries to the Cloud: Data is only sent to the cloud when using OpenAI's models. With Ollama, everything stays local.
Comparison to SmartConnections: Compared to other plugins, Smart2Brain offers full open-source access, supports local models, and provides a superior user experience.
Model Recommendations: While OpenAI models remain highly capable, for local embeddings, "mxbai-embed-large" has proven effective.
Multi-language Vaults: Supported, though response quality might vary based on the language model in use. OpenAI’s models generally perform best.
This comprehensive assistance tool aims to blend convenience with robust, offline AI-driven capabilities to redefine how users interact with their notes.