Lessons from building Agentic Memory, by CEO of Aperture Data, Vishakha Gupta-Cledat

the vibe

The agent memory space has come a long way in the last year. Graph structure is emerging, vector search is table stakes, and consolidation strategies like dream-state pruning and decay are actively being tried. The field is progressing. We spent time studying 20+ frameworks to understand where things stand, and one distinction kept surfacing that often gets collapsed: memory and cognition are not the same layer. Cognition is what models, agents, and humans do when they interpret, reason, and act. Memory is the infrastructure underneath that lets those actions persist, share, and evolve. This talk shares what we learned and what we built. Nexus is an open-source memory infrastructure and cognition hooks layer we've been designing around a Knowledge-Memory-Context framework. The hooks let the reasoning layer update, drop, enrich, and retrieve. The memory sits on ApertureDB, our unified vector, graph, and multimodal database, so it holds at production scale. We'll walk through the design decisions, show a live demo, and share the research behind it. Nexus is early stage and actively developing. We want your input on which memory problems matter most in what you're building. Bio: Vishakha Gupta-Cledat is the Co-founder and CEO of ApertureData, the company behind ApertureDB, a unified vector, graph, and multimodal database purpose-built for AI. ApertureDB is trusted by companies like Home Depot, Gracenote, Fanatics to power agentic and multimodal AI in production. She holds a…