A Personal Index is a private, always-updating knowledge graph stored on a user’s device or in a private cloud. It organizes personal data—like notes, messages, and files—so the Personal AI can better understand habits, preferences, and context, all while keeping data secure.

​
Personal Indexes provide

  • Smarter personalization: Personal Indexes organize user data, enabling on-device LLMs to recognize patterns and context for more relevant and adaptive responses.
  • Continuous improvement: Personal Indexes evolve in real time, continuously adapting to the user’s changing needs and preferences.

​
How do they work?

We use an on-device language model to turn raw documents into a knowledge graph. This approach (often called GraphRAG) works by:
  1. Extracting entities and relationships: The on-device language model reviews text and locates people, places, concepts, and how they connect.
  2. Indexing with graphs and vectors: Important facts are stored in a graph database for structured retrieval, and vector embeddings help the AI Agent quickly find relevant information.

​
Advantages

  • More accurate, context-aware answers.
  • Fewer incorrect responses (hallucinations).
  • Scalable for complex data without leaking privacy.