🎬 How to Permanently Fix Your Forgetful AI Agents (Full Guide)
Video Details
Channel: Arseny Shatokhin
URL: https://www.youtube.com/watch?v=ZcWMSVGcZio
Relevance: ⭐⭐⭐⭐
Summary
Comprehensive guide to implementing persistent agent memory: types of memory (episodic = conversation history, semantic = factual knowledge, procedural = skill memory), storage backends (PostgreSQL for structured history, vector DB for semantic search, Redis for fast cache), and LangGraph checkpoint integration for session persistence.
PUMA Relevance
Directly applicable to PUMA Stage 4–5. PUMA’s LangGraph uses PostgreSQL checkpoints for episodic memory (persisting the triage session state across issues in a sprint), Qdrant for semantic memory (historical issue similarity), and the sprint board state for procedural memory (which issues have been classified). The three memory types framework appears in PUMA’s architecture spec.