Startup Graphon raises $35M to give AI assistants a better memory

Graphon, a San Francisco startup, has raised $35 million to give AI assistants a better memory. Its software helps AI agents store and find information over time, so they are less likely to lose track of earlier conversations or miss how pieces of company information fit together.

The problem it targets

AI agents can write code, search databases and carry out multistep tasks. But they can still forget past conversations. They also struggle to make sense of huge company collections of information that keep changing.

How Graphon works

Graphon maps the connections among documents, videos and images. That includes where a piece of information came from and when it changed. CEO Arbaaz Khan compared common AI search systems to tearing pages out of a book, crumpling them up and tossing them in a bin. Searching that bin, he said, means hunting for pieces that each sit on their own.

Graphon instead builds a connected "fabric" of information. Khan says it works with off-the-shelf agents. "You can basically point your Claude or GPT or whatever to Graphon," he said. The name comes from "graphons," a math framework he used in Ph.D. research on coordinating swarms of robots.

Money and customers

• RRE Ventures led the $35 million Series A round.
• Energy Impact Partners, Hitachi Ventures, Gaia VC and HP also took part.
• The company raised about $8 million in a seed round earlier this year.
• The San Francisco 49ers and HP already use the software.

Graphon has 20 employees. It plans to spend the new money mainly on hiring engineers and researchers, plus some sales staff. It is aiming at individual developers and at companies with enormous stores of technical and operational data, such as industrial, financial services and healthcare firms. Khan said it will stay focused on businesses before moving into personal devices and consumer products.

What to watch

For any business planning to use AI agents, memory is a weak spot worth testing. A forgetful tool can give answers that miss context. The material does not say how well Graphon performs, what it costs, or whether outside tests back up its claims.