Graphon Labs builds a memory layer for AI agents that keeps relationships, sources, and timing across documents, images, audio, and video. As data arrives, the system builds a graph that discovers its own structure, with no hand-written ontology, using graphons, mathematical objects for very large graphs that co-founders Arbaaz Khan and Clark Zhang developed during robotics PhD research at the University of Pennsylvania. The product is for enterprises that need agents to reason over large operational archives, and for developers who want to point models they already use, including Claude and GPT, at that memory. It can run in Graphon's cloud, a customer VPC, on premises, or on a device. On November 10, 2026, Graphon announced a $35 million Series A led by RRE Ventures, with HP, Energy Impact Partners, Hitachi Ventures, and Gaia VC participating, less than six months after an $8.3 million seed. Total funding is now $43.3 million. The round is meant to move the company from early use into scaled enterprise deployments, open the infrastructure to developers, and extend the same memory onto devices, as agents start taking actions that depend on context outside any one window. The company reports the highest accuracy among systems it evaluated on LoCoMo (93.9 percent, using 58 percent fewer tokens than loading the full conversation), ConvoMem (90.2 percent), and ATM-Bench (81.7 percent, five points ahead of the next result on the public leaderboard), a University of Cambridge test of questions across video, audio, images, and text. It says the software is already deployed with the San Francisco 49ers, Hitachi, and other large enterprises. Khan is founder and CEO, Zhang is co-founder and CTO, and Deepak Mishra, formerly an AI investor, is co-founder and COO. Advisors include UC Berkeley professors Christian Borgs, a co-creator of graphon theory, and Jennifer Chayes.