MemMesh — persistent, self-improving memory for AI agents. Get started →
Get StartedConcepts at a Glance

Concepts at a Glance

This page is a quick tour of the vocabulary. Each idea has a dedicated page under Core Concepts.

Memory

A memory is a single durable unit of knowledge — a fact, a preference, a rule, or a decision. Memories are typed and timestamped, and they carry provenance so you can trace why the engine believes something.

Entity (subject)

An entity (also called a subject) is a canonical thing the memory graph tracks over time — a person, a project, an account. Subjects are the headline unit that MemMesh plans are sized around.

Scope

A scope controls who a memory belongs to and who can recall it — typically user, project, or platform. See Scopes.

The lattice

The lattice is the structured layer above raw memories: entities, their relationships, and the patterns mined from observed behavior. Prediction runs on the lattice. See Lattice & Patterns.

Prediction

MemMesh forecasts the next likely event for a subject and attaches a calibrated confidence score — one that is checked against what actually happened and recalibrated over time.

Lifecycle

The engine runs an observe → learn → recall loop. You feed it raw text; it decides what to keep and how to structure it; agents recall it later. See Lifecycle.