Knowledge Graph

Memoid’s knowledge graph extracts entities and relationships, enabling rich connected queries.

Concepts

Entities

Named things in your data:

• People — “John”, “Sarah Chen”
• Organizations — “Acme Corp”, “Google”
• Places — “San Francisco”, “New York”
• Concepts — “machine learning”, “project alpha”

Relationships

Connections between entities:

• John works_at Acme Corp
• Sarah manages John
• Acme Corp located_in San Francisco

Extracting Knowledge

With Memory Addition

Enable graph extraction when adding memories:

result = client.add(
    messages=[{
        "role": "user",
        "content": "My manager Sarah just got promoted to VP at TechCorp"
    }],
    user_id="user_123",
    extract_graph=True
)

# Extracted entities:
# - Sarah (person)
# - VP (role)
# - TechCorp (organization)

# Extracted relationships:
# - Sarah has_role VP
# - Sarah works_at TechCorp

Direct Extraction

Extract from any text:

result = client.extract_knowledge(
    text="John and Sarah co-founded Acme Corp in 2020. John is the CEO.",
    store=True  # Save to graph
)

print(result.entities)
# [John (person), Sarah (person), Acme Corp (organization), CEO (role)]

print(result.relationships)
# [John co-founded Acme Corp, Sarah co-founded Acme Corp, John has_role CEO]

Querying the Graph

Find Related Entities

result = client.graph_query(
    entity="John",
    depth=2  # Traverse 2 relationship hops
)

# Returns all entities within 2 hops of "John"
# John -> works_at -> Acme Corp -> located_in -> San Francisco

Search by Relationship

result = client.graph_search(
    query="Who works at Acme Corp?"
)

Managing Entities

Add Entity

client.add_entity(
    name="Acme Corp",
    type="organization",
    attributes={
        "industry": "technology",
        "founded": 2020
    }
)

Add Relationship

client.add_relationship(
    source="John",
    relation="works_at",
    target="Acme Corp"
)

Delete Entity

client.delete_entity("John")  # Also removes relationships

Entity Types

Common entity types:

• person
• organization
• location
• role
• project
• product
• event
• concept

Custom types are supported.

Use Cases

User Profile Building

Extract and connect information about users:

• Preferences
• Relationships
• Activities
• Context

Organizational Knowledge

Map company structures:

• Team hierarchies
• Project assignments
• Expertise areas

Recommendation Systems

Traverse connections:

• “People who like X also…”
• “Related to your network…”

Best Practices

1. Define your schema — Decide on entity and relationship types upfront
2. Handle ambiguity — “Apple” could be fruit or company
3. Merge carefully — Same entity may appear with different names
4. Prune regularly — Remove outdated relationships