Graphon Labs wins RRE's backing for an AI memory built on its founders' robot research

Graphon Labs, a San Francisco startup founded in July 2025 that builds memory software for AI agents, said on Tuesday, November 10, that it raised a $35 million Series A led by RRE Ventures, bringing its total funding to about $43 million, according to a company blog post.

The round came about six months after the company's seed announcement, the post said. Its company page puts total funding at $43.3 million.

A problem first seen in customer service

Chief executive Arbaaz Khan wrote in the post that, before starting Graphon, "I worked as an applied scientist on Amazon's Customer Service team." He described watching the industry move from intent classifiers to fine-tuned language models and then to retrieval, which feeds a model passages pulled from stored documents.

Retrieval, he wrote, breaks documents into chunks that can separate a rule from its exception. "'Can offer a refund' and 'can't offer a refund' concern the same topic and share almost all their words," he wrote, and the two can sit close together in the numerical space that search systems use.

From robot maps to company files

Khan and co-founder Clark Zhang, the chief technology officer, "spent the better part of our youth building graph-based solutions for robots," Khan wrote, helping them remember where they had been and navigate places they had not seen. The company says that research was done during their PhDs at the University of Pennsylvania's GRASP Lab. Deepak Mishra, the third co-founder, is chief operating officer.

The product builds a graph of how a company's videos, images, audio and documents connect, and uses a compressed mathematical summary of that graph, a graphon, to guide searches through it. "We spent the last year or so turning these ideas into a system that could ingest real data, build and maintain that structure, and answer questions over it at scale," Khan wrote. Developers reach it through an interface modeled on cloud file storage, with buckets, objects and queries, the post said.

Who uses it, and what the company claims

The company names two users on its website. "We're excited to explore how Graphon Labs' technology can help us gain a competitive edge," the San Francisco 49ers said in a statement on the company's site. Hitachi Rail said its evaluation "shows promise" in its industrial setting.

Graphon emerged from stealth in May with an $8.3 million seed round led by Novera Ventures, The Next Web reported. HP, Energy Impact Partners, Hitachi Ventures and Gaia VC joined the Series A, and the company has 20 employees, The Next Web reported, citing Business Insider. Graphon has not disclosed revenue or its valuation.

The company publishes benchmark results for its system. On one test of 500 questions over long chat histories, it says its deeper reasoning mode answered 481 correctly, or 96.2%. Those results are the company's own runs, though it posts the answers for each question.

The plan for the money

The Series A will go mainly to engineers and researchers, plus some sales staff, Khan told Business Insider, according to The Next Web. In the post, Khan wrote that Graphon will use the money "to bring this infrastructure to developers and enterprises," and later to cloud, on-premises and consumer devices.

A memory layer has to win against the model makers' own

Our reading is that Graphon's round buys time to turn benchmark wins into paying customers, which the company has not yet shown in numbers. Its strength is a technical idea with a clear origin in its founders' research, and two named users who say they are exploring or evaluating it, not that they have bought at scale.

The threat is that the largest model providers keep building memory into their own assistants, which would squeeze an independent layer. The opportunity is that companies that use several models want one memory that works with all of them. We would watch for disclosed revenue or a named production customer by the next round; our view would change if one appears.

The view in this final section is FinanceInTime's editorial opinion, for information only. It is not investment, tax, legal or other advice, and it does not take your circumstances into account. Forecasts are uncertain and past results do not guarantee future ones.