Graphon Raises $35 Million for an AI Agent Memory Layer

James Calloway

Graphon, a San Francisco startup, has raised $35 million in a Series A round to expand software that lets AI agents store and retrieve company information over time. Its data-mapping system connects documents, videos, and images by their relationships, origin, and changes rather than treating them as isolated search snippets.

RRE Ventures led the round, joined by Energy Impact Partners, Hitachi Ventures, Gaia VC, and HP. The Series A followed a roughly $8 million seed round closed earlier in 2026.

What problem is Graphon trying to solve?

AI agents already write code, search databases, and complete multistep tasks. But they can lose track of earlier conversations while their users face large, constantly changing collections of company documents and media. Graphon’s software gives those agents a data layer designed to preserve context across sessions.

The company takes its name from “graphons,” a mathematical framework that founder and CEO Arbaaz Khan used during his Ph.D. research into coordinating swarms of robots. Graphon applies a graph-based approach to information: it maps links among records, captures where each piece came from, and tracks when it changed.

Khan described the common search model as tearing pages from a book, crumpling them, and throwing them into a bin. He told Business Insider: “Now when you go searching that trash can, you’re just trying to find the pieces to answer. You’ve basically lost all that information because each piece is living on its own.”

Graphon instead calls its connected structure a “fabric” of information. In practice, the product retains relationships between items instead of flattening them into independent matches. That structure can give an agent the provenance and chronology needed to interpret a result.

Which workflows does Graphon target?

Graphon identifies two customer groups:

• Individual developers.
• Companies with enormous stores of technical and operational data, including industrial businesses, financial-services firms, and healthcare companies.

The integration model does not require customers to replace their agent. “You can basically point your Claude or GPT or whatever to Graphon,” Khan said. “You can push in any amount of data in this, and we literally mean any.” Graphon says its software works with any off-the-shelf agent.

The company already lists the San Francisco 49ers and HP as users. At the interface layer, an agent can retrieve connected information spanning documents, video, and images rather than rely on isolated excerpts. The announcement provides no accuracy, latency, or cost measurements, so the benefit remains unquantified in this funding disclosure.

Where will the money go?

Graphon has 20 employees. It plans to direct the proceeds primarily toward hiring engineers and researchers, with some sales and go-to-market staff added. The company expects to remain focused on enterprise customers before moving further into personal devices and consumer products.

Graphon also published the pitch deck it used to secure the funding. Once hiring begins, its practical test will be whether connected retrieval improves agent accuracy and continuity without slowing responses, while retaining compatibility with off-the-shelf agents.

The $35 million gives Graphon the staff and time to test whether relationship-aware retrieval can replace isolated search without adding agent delay.