Graph Memory

AgentRecall builds a Neo4j-powered knowledge graph from your memories, automatically extracting entities and relationships.

How It Works

When you store a memory, the AI processing pipeline extracts:

• Entities — People, places, concepts, objects mentioned in the text
• Relationships — How entities connect to each other (e.g., "works_at", "prefers", "is_allergic_to")
• Properties — Attributes of each entity (e.g., name, type, value)

These are stored as nodes and edges in Neo4j, forming a rich graph structure.

Example Graph

[User] --prefers--> [Dark Mode]
[User] --allergic_to--> [Peanuts]
[User] --works_at--> [Acme Corp]
[Acme Corp] --located_in--> [San Francisco]
[User] --uses--> [Python] --is_a--> [Programming Language]

Traversal Queries

Use traverse() to walk the graph and find connections between entities.

Python

# Find all entities connected to a node
connections = memory.traverse(
    entity="User",
    depth=2,
    agent_id="my-agent"
)

for node in connections:
    print(f"{node.entity} --{node.relationship}--> {node.target}")

Node.js

const connections = await memory.traverse({
  entity: "User",
  depth: 2,
  agentId: "my-agent",
});

connections.forEach(node => {
  console.log(`${node.entity} --${node.relationship}--> ${node.target}`);
});

Finding Hidden Connections

Graph traversal can discover non-obvious links between concepts. For example:

# "What connects the user to San Francisco?"
paths = memory.traverse(
    source="User",
    target="San Francisco",
    agent_id="my-agent"
)
# Result: User -> works_at -> Acme Corp -> located_in -> San Francisco

Graph via Cloud API

# POST /v1/traverse
curl -X POST https://api.agentrecall.cloud/v1/traverse \
  -H "Authorization: Bearer *** \
  -H "Content-Type: application/json" \
  -d '{
    "entity": "User",
    "depth": 3,
    "agent_id": "my-agent"
  }'