zachrizzo/sonder-memory

Sonder-memory attempts to solve the 'GraphRAG' problem (retrieval augmented generation over structured graphs) using an information-theoretic approach (Shannon entropy convergence). While the theoretical approach of using entropy to prune or guide graph diffusion is a novel combination of disciplines, the project currently sits at zero stars and zero forks, indicating it is in a nascent or personal experiment phase. It faces massive competitive pressure from established players like Microsoft (GraphRAG), Neo4j (which is aggressively integrating LLM features), and specialized startups like WhyHow or FalkorDB. Frontier labs like OpenAI and Google are also treating 'agentic memory' and structured retrieval as core platform capabilities, making any independent KG-memory project highly susceptible to displacement. The lack of an ecosystem or data gravity means its only moat is the specific algorithm, which—if successful—would likely be absorbed or reimplemented by larger open-source frameworks like LangChain or LlamaIndex within months.