RRE Leads Graphon Labs' $35M Series A for a Graph-Based Memory Layer That Lets Agents Keep Context, Sources and Timing

Graphon Labs, a San Francisco company that builds a memory layer for AI agents, has raised a $35 million Series A. It announced the round on November 10, 2026. RRE Ventures led it, with HP, Energy Impact Partners, Hitachi Ventures and Gaia VC also taking part. The raise brings Graphon’s total funding to $43.3 million. The company says the round comes less than six months after it emerged from stealth with an $8.3 million seed. The announcement does not disclose a valuation.

Graphon calls the product Persistent Intelligence. It is meant to let agents keep what they have read, seen or heard, along with how those pieces of information connect, where they came from and when they changed.

A connected map instead of torn-up pages

Graphon says common retrieval methods, including retrieval-augmented generation (RAG), GraphRAG, context windows and traditional search, can retrieve information but compress or lose relationships, history and timing as data grows and changes. Its alternative is built on graphons, a mathematical framework for analyzing very large graphs. The company says its founders developed that research during their PhDs at the University of Pennsylvania’s GRASP Lab.

“The world is heading toward agents and physical systems that create and depend on more information than anyone has ever had to manage,” said Arbaaz Khan, Graphon’s founder and CEO. “In a world where we are moving to AI using computers, we need to fundamentally rethink how computers access data to pre-fill their context windows. The old way, using vector search, is like searching through the pages of a torn-up book. We take those pages and turn them into a connected map, unlocking seamless memory search.”

The architecture, Graphon says, rethinks storage, search, retrieval, reasoning and memory so that what an agent knows, what it has done, what changed and what happened as a result stay linked, with sources and timing intact. The company also says the approach cuts the compute time needed to represent and reason across large amounts of information. That compute claim is Graphon’s own and is not tied to a published benchmark in the announcement.

Graphon’s own scores on LoCoMo, ConvoMem and ATM-Bench

Graphon says it recorded the highest accuracy among the systems it evaluated on three long-term context tests: LoCoMo, ConvoMem and ATM-Bench. All three figures below are the company’s own results.

On LoCoMo, Graphon reports 93.9% accuracy using 58% fewer tokens than loading the full conversation into the model. On ConvoMem, it reports 90.2% accuracy with 95% fewer tokens. On ATM-Bench, a University of Cambridge benchmark for question answering across video, audio, images and text, it reports 81.7%, which it says is 5.0 percentage points ahead of the next system on the public leaderboard. The company says the LoCoMo and ConvoMem comparisons use baseline results published by Databricks.

Graphon’s website currently lists slightly higher figures for two of those tests: 94.0% on LoCoMo and 91.0% on ConvoMem. The 81.7% ATM-Bench score from the announcement does not appear as a headline number on the site, which instead says Graphon holds the top two places on the public leaderboard.

49ers and Hitachi Rail, plus a path onto devices

The company says it is already deployed with the San Francisco 49ers, Hitachi and other large enterprises. Costa Kladianos, the 49ers’ executive vice president of technology, said the team is “excited to explore” how the technology can help, including with sponsorship opportunities. Amaury Jourdan, deputy manager and CTO for digital and AI activities at Hitachi Rail, described an evaluation that “shows promise” for discovering context knowledge, connecting data sources and improving AI inferencing in that industrial setting.

The same architecture, Graphon says, underpins three products: Graphon Enterprise, a governed memory layer over enterprise data that can be deployed on hosted infrastructure, a Graphon-managed or customer-managed virtual private cloud, or on-premises systems; Graphon Developer, for builders who do not want to recreate a context stack; and Graphon On-Device Memory, for phones, laptops, cameras, wearables and other devices.

“This funding takes us from early experimentation to scaled enterprise deployments, opens our infrastructure to developers, accelerates our research, and brings Persistent Intelligence on-device,” Khan said.

Raju Rishi, a general partner at RRE Ventures, said models can reason and agents can act, but neither improves if they cannot build on prior knowledge. He said the firm led the round because it sees Graphon becoming “the source of truth beneath every agent.”

Graphon was founded in July 2025 by Khan, chief technology officer Clark Zhang and chief operating officer Deepak Mishra.