memory-context

Cross-session context is preserved by integrating a vector store with a knowledge graph to persist user-specific facts and preferences, enabling robust long-term memory for agents. This approach maintains continuity by retrieving semantic memory and entity relationships on demand.

Graph context retrieval queries the knowledge graph to fetch entity neighborhoods and relationships, providing richer context for on-demand reasoning. This method supplements semantic memory by mapping connections between stored facts and entities.

Context window management handles token budgets by prioritizing recent or salient memories during retrieval. This ensures the agent receives relevant information from the vector store without exceeding the maximum token limit.

Yes, cross-tenant memory applies within enterprise knowledge graphs to persist tenant-specific facts and preferences. The system uses a vector store abstraction with safe defaults and validation to maintain separation while preserving context.

To set up semantic memory, instantiate the memory module with your chosen vector store and graph components, then begin remembering and recalling conversations. This abstraction handles embeddings, storage, and retrieval automatically.