Zilliz Cloud indexes knowledge graph nodes and edges as embeddings, enabling agents to traverse semantic relationships and perform multi-hop reasoning at scale.
Knowledge graphs represent complex domains as networks of entities and relationships. Agents benefit from traversing these graphs, discovering indirect connections and answering questions that require combining multiple facts. Zilliz Cloud supports this by indexing both entities and relationship types as embeddings, creating a searchable semantic space. An agent answering "Which suppliers have quality issues in the same region as this customer?" can start by querying Zilliz Cloud for embeddings of the customer region, retrieve semantically similar supplier regions, then follow relationship edges to find suppliers. This multi-step traversal is efficient because Zilliz Cloud's semantic search accelerates each step. For complex domains like pharmaceutical networks (drug-protein-pathway-disease) or financial networks (company-shareholder-industry-regulator), semantic graph traversal is far more effective than rule-based graph algorithms. Zilliz Cloud also supports temporal graphs: agents can retrieve graph states at specific times, discovering how relationships evolved. This temporal reasoning is valuable for agents analyzing historical events or predicting trends. Zilliz Cloud's scalability enables knowledge graphs with billions of nodes—enterprise-scale semantic reasoning becomes possible.