Skip to main content
CORE’s memory graph is the foundation of how your digital brain works. Unlike traditional databases that store isolated facts, CORE creates a living network of connected information.

​
How Your Memory is Structured

Your memory consists of three types of building blocks:
  1. Episodes - The raw conversations and interactions you have
    • Every chat with Claude, message in Slack, or note you add becomes an episode
    • Episodes preserve the original context and serve as the source of truth
  2. Entities - The people, places, concepts, and relationships in your world
    • Names like “Manik,” concepts like “React,” companies like “TaskMaster”
    • Even relationships like “works at” or “prefers” are treated as entities
  3. Statements - The specific facts extracted from your episodes
    • “Alex works on TaskMaster” or “Manik prefers TypeScript”
    • Each statement knows when it became true and links back to its source episode

​
Why This Structure Matters

This approach gives CORE unique advantages over simple note-taking or search systems:
  • Traceable Knowledge: Every fact in your memory can be traced back to the original conversation where you mentioned it
  • Smart Connections: When you mention “TaskMaster,” CORE finds not just that word, but all related people, timelines, and decisions connected to your project
  • Evolving Understanding: As you have more conversations, CORE builds richer connections between existing entities rather than creating isolated notes

​
How Facts Are Organized: The 11 Aspects

CORE doesn’t just extract facts—it categorizes them. Every statement is classified into one of 11 aspects:
  • Identity: Who you are → “Manik works at Red Planet”
  • Preference: How you want things → “Prefers TypeScript over JavaScript”
  • Decision: Choices made → “Chose Prisma for ORM”
  • Directive: Hard rules → “Always run tests before PR”
  • Knowledge: What you know → “Expert in React”
  • Problem: Challenges faced → “Hit rate limits with GitHub API”
  • Goal: What you’re working toward → “Launch MVP by Q2”
  • And 4 more: Belief, Action, Event, Relationship
When Claude needs your coding preferences, CORE doesn’t search everything—it filters by aspect = Preference. When debugging, it surfaces aspect = Problem. This is how CORE provides the right information at the right time. See Statement Aspects for detailed examples of each aspect.

​
How Recall Works: Intent-Driven Retrieval

When you search your memory, CORE classifies your query into one of 5 types:
  1. Aspect Query: “What are my preferences?” → Filter by aspect
  2. Entity Lookup: “Tell me about Sarah” → Traverse entity graph
  3. Temporal Query: “What happened last week?” → Filter by time
  4. Exploratory: “Catch me up” → Recent summaries
  5. Relationship Query: “How do I know Sarah?” → Multi-hop traversal
Traditional RAG: Same search for all queries (slow, imprecise) CORE: Different strategy per query type (fast, precise) See Query Types for detailed examples and performance comparisons.