Memory that sees patterns: inside Ari's temporal knowledge graph

Most assistant memory is a list of notes. Notes do not know how they relate to each other, and they do not know when they stopped being true, so an old note and a new one can sit side by side and contradict each other.

Ari's memory is a knowledge graph: people, preferences, goals, health notes, events, projects, places, and habits are stored as entities, and the facts between them as connections. "Sam is your sister" and "Sam is visiting in May" are linked through Sam, so asking about the visit brings the relationship with it.

The graph is temporal. Each fact records when it became true and, when something changes, when it stopped. Ari sorts facts into current, historical, and replaced, so "I moved to Calgary" makes the old city history instead of erasing it. Ari can answer what is true now and what used to be true.

The engine underneath is Graphiti, an open-source library for building temporal knowledge graphs from conversations. Ari adds its own types on top, including people, goals, preferences, health, events, projects, habit opportunities, and behaviour patterns, so the graph holds the things that matter for follow-through.

Every day Ari looks back over your recent conversations and records behaviour patterns with the evidence behind them: a habit that slips on the same weekday, a plan that keeps colliding with late shifts, sleep that gets shorter in deadline weeks.

Those patterns shape the next check-in. Ari brings a pattern up when it helps you act on it, and you can ask Ari directly what Ari has noticed about your sleep, routines, or plans.

The memory explorer in the web app shows what Ari remembers as a map. Search it, filter by type and date, and see replaced facts separately from current ones.

If something is wrong, tell Ari and Ari corrects it. Ask Ari to forget something and Ari stops using it; the forgotten fact is kept only as history, never as something Ari acts on. To erase everything, email support and the account is deleted.

Your chats are never used to train AI models, and your data is never sold.