Semantic Memory

When to Use

Use semantic memory for facts, concepts, and relationships. Unlike episodic memory (which is about "events"), semantic memory is about "knowledge". Perfect for:

• User profiles (Name, Occupation, Family)
• World knowledge (Product details, Policies)
• Extracted entities and their relationships

Key Operations

Code Example

Here we manually add facts extracted from a conversation. In a real automated agent, these might be extracted by an LLM first.

# 1. Add extracted facts to the Knowledge Graph
facts = [
    "Marcus is a senior engineer at Emma's company",
    "Emma has a sister named Sarah",
    "Nana Rose is Emma's grandmother"
]

for content in facts:
    client.memory.facts.add_fact(
        content=content,
        kg_name="KG_Companion_Demo",
        metadata={"category": "relationships"}
    )

# 2. Query the Knowledge Graph
# The system can answer questions by synthesizing these facts
results = client.memory.facts.search(
    query="Who are Emma's family members?",
    kg_name="KG_Companion_Demo"
)

for res in results:
    print(f"Fact found: {res.content}")

Key Takeaways

• Semantic memory consolidates information. If you add "Emma is a PM" and later "Emma works as a Product Manager", the system handles the semantic similarity.
• Facts are stored in a Knowledge Graph, allowing for structured retrieval.
• This is the "long-term knowledge base" of your agent.