MemoryNode¶

Represents an entity in the knowledge graph.

Structured Data Import

You can directly create and add custom entities without LLM extraction. See the Structured Data Guide for complete examples.

Definition¶

@dataclass
class MemoryNode:
    id: str
    name: str
    entity_type: str
    description: str = ""
    embedding: list[float] = None
    importance: float = 1.0
    access_count: int = 0
    last_accessed: datetime = None
    created_at: datetime = None
    user_id: str = "default"
    memory_id: str = "default"
    aliases: list[str] = None
    properties: dict = None

Fields¶

Field	Type	Description
`id`	str	Unique identifier
`name`	str	Entity name
`entity_type`	str	Type (Person, Company, etc.)
`description`	str	Entity description
`embedding`	list[float]	Vector embedding
`importance`	float	Importance score (0-1)
`access_count`	int	Number of times accessed
`last_accessed`	datetime	Last access time
`created_at`	datetime	Creation time
`user_id`	str	Owner user ID
`memory_id`	str	Memory context ID
`aliases`	list[str]	Alternative names
`properties`	dict	Additional properties

Example¶

from graphmem.core.memory_types import MemoryNode, MemoryImportance
from datetime import datetime

node = MemoryNode(
    id="elon_musk_001",
    name="Elon Musk",
    entity_type="Person",
    description="CEO of Tesla and SpaceX",
    importance=MemoryImportance.CRITICAL,  # Key person, never decays
    aliases={"Musk", "Elon", "E. Musk"},   # Set for alias matching
    properties={
        "title": "CEO",
        "companies": ["Tesla", "SpaceX", "Neuralink"],
    },
    user_id="alice",
    memory_id="chat",
)

Adding to GraphMem¶

from graphmem import GraphMem, MemoryConfig

# Initialize
memory = GraphMem(config, memory_id="my_agent", user_id="default")
memory._ensure_initialized()

# Add custom entity
memory._memory.add_node(node)
memory.save()

# Now you can query it!
response = memory.query("Who is the CEO of Tesla?")

Methods¶

to_dict()¶

Convert to dictionary.

data = node.to_dict()

from_dict()¶

Create from dictionary.

node = MemoryNode.from_dict(data)