trilium/docs/User Guide/AI Integration/01-overview.md

AI Integration Overview

Introduction

Trilium Notes integrates advanced AI and Large Language Model (LLM) capabilities to enhance your note-taking and knowledge management experience. These features enable intelligent search, content generation, and interactive assistance while maintaining control over your data.

What AI Features Are Available

Core Capabilities

1. Chat with Notes

An interactive AI assistant that can:

• Answer questions based on your note content
• Provide summaries of complex topics across multiple notes
• Help you discover connections between ideas
• Assist with research and analysis

2. Semantic Search

Advanced search capabilities using embeddings:

• Find conceptually related notes even when exact keywords don't match
• Discover hidden connections between notes
• Retrieve relevant context automatically during conversations

3. Tool-Enabled Actions

AI can perform actions within your knowledge base:

• Create new notes based on conversations
• Update existing note content
• Search and retrieve specific information
• Manage note attributes and relationships
• Work with calendar and date-based notes

4. Content Generation and Enhancement

• Generate summaries of long notes
• Expand on ideas and concepts
• Help with writing and editing
• Create structured content from unstructured information

Supported AI Providers and Models

Cloud Providers

OpenAI

• Models: GPT-4, GPT-4 Turbo, GPT-3.5-turbo
• Embedding Models: text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002
• Advantages:
  • Most widely tested and stable
  • Excellent general knowledge
  • Strong reasoning capabilities
• Considerations:
  • Requires API key and internet connection
  • Usage-based pricing

Anthropic

• Models: Claude 3 Opus, Claude 3 Sonnet, Claude 3 Haiku
• Advantages:
  • Strong analytical capabilities
  • Excellent for complex reasoning tasks
  • Good context handling for long documents
• Considerations:
  • Requires API key and internet connection
  • Usage-based pricing

Local/Self-Hosted Options

Ollama

• Models: Various open-source models (Llama 3, Mistral, Phi, etc.)
• Embedding Models: mxbai-embed-large, nomic-embed-text, all-minilm
• Advantages:
  • Complete privacy - runs entirely on your machine
  • No API costs
  • Full control over model selection
  • Works offline
• Considerations:
  • Requires local compute resources
  • May be slower than cloud providers
  • Model quality varies

Use Cases for AI in Note-Taking

Research and Learning

• Literature Review: Summarize and synthesize information from multiple research notes
• Concept Exploration: Ask questions to understand complex topics from different angles
• Study Assistant: Create practice questions, summaries, and explanations from study notes

Knowledge Management

• Information Retrieval: Quickly find specific information across thousands of notes
• Connection Discovery: Identify relationships between seemingly unrelated notes
• Content Organization: Get suggestions for better organizing and structuring your notes

Writing and Content Creation

• Idea Development: Expand on initial thoughts and brainstorm new directions
• Draft Generation: Create first drafts based on outline notes
• Editing Assistance: Improve clarity, structure, and flow of existing content

Personal Productivity

• Meeting Summaries: Generate concise summaries from meeting notes
• Task Extraction: Identify action items from project notes
• Daily Reviews: Get AI-generated summaries of daily activities and progress

Creative Projects

• Story Development: Explore plot ideas and character development
• World Building: Maintain consistency in creative universes
• Idea Synthesis: Combine concepts from different notes to generate new ideas

Privacy and Security Considerations

Data Control

• Local Processing Options: Use Ollama for complete privacy with no data leaving your machine
• Note Exclusion: Mark sensitive notes with #excludeFromAI label to prevent AI access
• Provider Selection: Choose between cloud and local providers based on your privacy needs

Security Features

• API Key Protection: Encrypted storage of API keys
• Session Isolation: Each chat session is isolated and temporary
• No Automatic Uploads: Notes are only sent to AI providers when explicitly requested

Best Practices

1. Review Provider Policies: Understand how each provider handles your data
2. Use Local Models for Sensitive Data: Consider Ollama for confidential information
3. Label Sensitive Notes: Use the exclusion label for notes that should never be processed by AI
4. Regular Audits: Periodically review which notes are being accessed by AI features

Getting Started

To begin using AI features in Trilium:

1. Enable AI Features: Navigate to Options → AI/LLM and enable the feature
2. Choose a Provider: Select between OpenAI, Anthropic, or Ollama based on your needs
3. Configure Embeddings: Set up embedding generation for semantic search
4. Start Chatting: Use the "Chat with Notes" button to begin interacting with your knowledge base

Architecture Overview

The AI integration in Trilium follows a modular architecture:

Three-Layer System

1. Provider Layer: Interfaces with different AI providers (OpenAI, Anthropic, Ollama)
2. Service Layer: Manages conversations, context, and tool execution
3. Integration Layer: Connects AI capabilities with Trilium's core features

Key Components

• Chat Service: Manages conversations and message history
• Context Service: Extracts and formats relevant note content
• Tool Registry: Enables AI to perform actions within Trilium
• Embedding Service: Generates and manages semantic embeddings
• Model Capabilities: Tracks and manages different model features

This architecture ensures flexibility, allowing you to switch providers or models while maintaining consistent functionality across the application.