GRAPHON RAISES $35M TO BUILD LONG-TERM MEMORY FOR AI AGENTS Traditional RAG systems split data into chunks and retrieve similar passages, often losing relationships between facts, their original sources, and how information changes over time. Graphon takes a different approach. It builds a self-supervised, schema-free knowledge graph across documents, images, and videos, allowing AI agents powered by GPT, Claude, or other models to access persistent, connected information. The startup reports the following benchmark results: • ATM-Bench: 81.7% (#1#) • LoCoMo: 94% • LongMemEval: 96.2% • DolphinBench: 84%, reportedly at 10x lower cost Graphon says its system can also ingest 100K PDFs without redesigning its underlying data schema. The $35M Series A was led by RRE Ventures, with participation from HP and Hitachi Ventures. Graphon says HP and the San Francisco 49ers already use its technology.