memory-systems

Persistent agent memory is implemented using layered architectures ranging from working memory to temporal knowledge graphs. This approach provides strategies for retaining state and knowledge across multiple interactions, overcoming the context loss that occurs between sessions.

Simple vector stores for agent memory lack temporal context and structural relationships. This skill addresses these limitations by providing advanced patterns like knowledge graphs and vector RAG with metadata to capture complex, time-aware state.

Building a temporal knowledge graph for agent state involves structuring data with time-based relationships to track how knowledge evolves. This skill provides implementation patterns and benchmarking guidelines for designing these advanced memory architectures.

Memory consolidation in a persistent agent system is best handled by summarizing and structuring data across different memory layers. The skill provides guidelines for consolidating working memory into long-term knowledge graphs to optimize retrieval performance.

Yes, a file-system-as-memory approach can be used for cross-session persistence. This skill covers this implementation pattern alongside vector RAG with metadata and knowledge graphs to provide flexible options for retaining agent context.