skill-system-memory

Persistent memory for AI agents is added by connecting to a PostgreSQL database to store and retrieve past experiences. This Skill provides a multi-type memory system that handles working, episodic, semantic, and procedural data to prevent intelligence loss.

You can manage four distinct memory types: working, episodic, semantic, and procedural memory. Each type has distinct lifespans and uses, supported by decay and pruning mechanisms to manage data retention over time.

Agent memory search in PostgreSQL uses a hybrid approach combining full-text search, trigram similarity, and tag filtering. It also optionally supports pgvector for semantic search to ensure precise retrieval of stored experiences.

Yes, you can track ongoing tasks using the integrated task management layer. This feature allows you to track and prioritize work directly linked to the stored memories, managing related workflows within the same system.

No, you do not need pgvector to use the persistent agent memory system. pgvector is an optional dependency used specifically for enabling semantic search, while standard full-text and trigram search work independently.

Memory overload in a persistent AI agent system is prevented using built-in decay and pruning mechanisms. These features automatically manage the lifespan of working, episodic, semantic, and procedural memories to optimize storage.