Skip to content

The memory graph

Most assistants keep memory as a pile of notes they skim. Moko turns everything it learns into a graph of subjects and how they connect — knowledge it’s fluent in, not text it re-reads.

  • Subjects — the things in your world: people, places, organizations, objects, events, topics.
  • Attributes — typed facts about a subject: “Maya’s birthday is March 14.”
  • Relationships — typed, directional links: “Maya is your sister,” “Tom is Maya’s husband.”
  • Sources — where each fact came from: the message, the page, the moment.
  • Changes supersede, they don’t erase. A new value replaces the old one with the history kept.
  • So you can ask backwards. “What did I used to think about that?” still has an answer.

Because the graph is structured, Moko answers precise questions:

  • “When did I last talk to Sarah?”
  • “Who is Maya’s husband?”
  • “Where does my brother live?” — even when it has to hop you → your brother → his city.

When you chat, Moko pulls the relevant slice into the conversation, so it answers from memory — not from a generic guess.

  • It folds aliases together. “My sister,” “Maya,” and “Maya Tanaka” become one subject.
  • Safe merges happen instantly. The uncertain ones get flagged and reconciled overnight during dreaming, with full context and no rush.
  • It records how you seemed — perceived emotion plus how confident it is — held gently as a hypothesis, separate from its own feelings.
  • That’s time-stamped too, so trends become legible: “you’ve seemed flat this week” is a pattern Moko can actually notice.
  • The live chat is working memory — a small, fresh window.
  • The graph is durable memory. Everything that matters is extracted as you talk.
  • So a fresh session loses nothing. Moko drops the verbatim back-and-forth, never the knowledge.