conversation-memory

Persistent memory for LLM conversations is maintained by applying a tiered architecture that stores and retrieves short-term, long-term, and entity-based memory across sessions. This preserves user preferences, entities, and context over time.

Short-term memory retains immediate context during an active chat, while long-term memory preserves user preferences and facts across distinct sessions. Entity memory specifically tracks facts about users and referenced entities.

Memory retrieval surfaces relevant stored memories by applying memory-aware prompting within the LLM conversation. This mechanism injects preserved context and user details directly into the active prompt.

Yes, the memory lifecycle management includes built-in privacy and safety safeguards for user data. Safe memory controls are applied when configuring memory scopes for individual users with consent.

Entity-based memory is needed when a knowledge-driven agent must track specific facts about users and referenced entities over time. It preserves continuity and relevance by applying memory-type definitions within a tiered architecture.