How AI remembers.

Retrieval is an event. Memory is a process.

Most retrieval systems answer a question by finding nearby text. That is useful, but it does not explain what should persist, what should change, or how a later correction should reshape future recall.

EngramRAG begins from a different premise: memory has a lifecycle. Material is formed into candidate knowledge, consolidated into relationships, recalled through context, and revised by people who can see the evidence behind it.

Four moments of memory

1. Formation — documents, data, decisions, and code enter with their original location intact.
2. Consolidation — repeated concepts and relationships are reconciled into a connected ontology.
3. Associative recall — the system retrieves through entities, relationships, time, and provenance rather than lexical similarity alone.
4. Human revision — a person can correct the model of the world without erasing the history that led there.

Provenance is part of the answer

A remembered fact without its origin is difficult to inspect. Consilience keeps the path from source to entity to relationship to action visible. The trail gives a team a way to review, revise, and verify agent work.

RAG retrieves. EngramRAG remembers.

EngramRAG is the memory architecture and retrieval methodology inside Consilience. It is inspired by how humans remember, but built for how AI systems actually work.