Multi-Hop Search

Use graph-aware retrieval to answer questions that require traversing connected entities across memories.

Scenario

You need answers that span multiple steps, for example:

• customer -> recent issue -> impacted product -> owner.

Step 1: Ingest Connected Data

client.memory.add(
    content="Customer Acme reported login timeout in Product Portal. Owner is Team Identity.",
    external_user_id="ops_user_001",
    memory_policy={"mode": "auto"}
)

Step 2: Run Agentic Search

response = client.memory.search(
    query="Which team owns the product affected by Acme login timeout?",
    external_user_id="ops_user_001",
    enable_agentic_graph=True,
    max_memories=20,
    max_nodes=20
)

Step 3: Use Node and Memory Outputs Together

Combine:

• memory snippets for evidence
• nodes/relationships for structured traceability

Search Tuning Checklist

• Keep enable_agentic_graph=true
• Increase max_nodes for deeper traversals
• Use explicit schema when your domain is highly structured

Next Steps

• Advanced Search Guide
• Knowledge Graphs Guide