Agentic Retrieval

Agentic retrieval uses graph-aware memory search to improve relevance for ambiguous or multi-step queries.

When to Use

• User questions contain implicit references.
• Answers require connecting multiple entities.
• Vector-only retrieval misses business context.

Recommended Baseline

{
  "enable_agentic_graph": true
}

Query params:

• max_memories=20
• max_nodes=15
• response_format=toon for LLM-bound contexts

Example

results = client.memory.search(
    query="What blockers affect Acme onboarding and who owns them?",
    external_user_id="pm_001",
    enable_agentic_graph=True,
    max_memories=20,
    max_nodes=20
)

Troubleshooting

• Too shallow: increase max_nodes.
• Irrelevant scope: set organization_id and namespace_id.
• Noisy query: narrow query intent and include key entities.

Related

• Retrieval
• Search Tuning
• Multi-Hop Search Tutorial