KNOWLEDGE GRAPH

Memory that's all connected

Agent Memory stores memory as a knowledge graph: entities as nodes, typed relationships as edges. And because edges are documents, they carry their own properties, confidence, and timestamps - so the connections are as queryable as the facts themselves.

Join the waitlist Technical deep-dive

01 |CONNECTED MEMORY

Entities as nodes, relationships as edges

People, systems, facts, and decisions become nodes; how they relate becomes typed edges you can traverse. One question can walk from a person to their team to the decisions they own, in a single step of graph traversal - no join, no second system.

chunk

embedding vector

text

extracted_from

attribute

text

confidence

valid_from

has_source

source

kind

trust

belongs_to

entity

type

name

embedding vector

relation

kind (verb)·valid_from·valid_until

entity

type

name

embedding vector

A graph, not a pile of vectors

Relationships are first-class, queryable structure - so memory carries meaning, not just proximity in vector space.

Typed relationships

Edges name how things relate, so the graph carries meaning, not just similarity.

Learn more

Edges are documents

Each edge carries its own properties, confidence, and timestamps - queryable like any record.

Learn more

Traversal in one step

Walk from entity to entity to any depth, without a join or a separate graph store.

Learn more

Vectors on the same nodes

Embeddings sit on the graph, so similarity and relationships are ranked together.

Learn more

Facts stay linked

New memories attach to the entities they're about, so context compounds instead of fragmenting.

Learn more

One copy of the data

The graph, the documents, and the vectors are the same records - nothing to sync.

Learn more

02 |ONE SUBSTRATE

Relationships you can't get from a vector store

A flat vector store can tell you two memories read similarly. It can't tell you that this person owns that decision, or that this fact superseded that one. Agent Memory keeps the connections as first-class, queryable structure.