memory-systems

To build persistent cross-session memory for AI agents, you need an architecture that stores and retrieves facts, entities, and temporal relationships. This skill provides frameworks and patterns to implement backends like vector stores and knowledge graphs for continuous context.

Storing temporal knowledge graphs for agent memory involves structuring entities and relationships over time. This skill provides architectural patterns and practical examples for querying temporal graphs to help agents reason over evolving facts across conversations.

Mem0, Zep/Graphiti, Letta, Cognee, and LangMem differ in their retrieval strategies and memory consolidation patterns. This skill evaluates these frameworks with up-to-date benchmarks to help you select the right backend for your agent's scaling requirements.

Yes, you can use a vector store to retain user preferences across AI conversations. This skill provides code snippets for integrating vector store backends, enabling your agents to retrieve and reason over stored user facts and preferences seamlessly.

Relying on a single framework for agent memory can limit your ability to handle complex temporal relationships or scale retrieval effectively. This skill defines layered memory architectures and consolidation patterns to address structural limitations and improve cross-session reasoning.