Stop Using RAG as Memory
summary
Argues that vectorstore RAG fails as memory and shows a working alternative: temporally-aware knowledge graphs built with the open-source Graphiti/Zep stack. Demonstrates SDK code (TypeScript, Python, Go), developer-defined business-object entities (financial goals, debts, income) with validation (pydantic/zod/go structs), a live demo storing a $5,000 rent node, concurrent searches and the ability to filter by node types, plus provenance and temporal context that prevent irrelevant semantic hits. Links to a paper for details.
AI-written summary · about
related talks
27mLies, Damn Lies, and AIs
27mWhen Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
30mAgentic RAG with Knowledge Graphs Explained
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
36mAgentic GraphRAG: AI’s Logical Edge
38mLeveraging Knowledge Graphs for RAG: A Smarter Approach to Contextual AI Applications