AI memory startup focused on cutting token costs raises $98 million

As corporate America finally begins to crack down on developers’ use of AI, an 8-month-old startup called Engram sees a big business opportunity in helping companies save money.

Engram announced Tuesday that it has raised $98 million from investors including General Catalyst, Kleiner Perkins and Sequoia, as well as OpenAI co-founder Andrej Karpathy, who recently joined Anthropic.

The startup, which calls itself the “learned memory” of AI, says its models can anticipate questions and provide smarter answers with cheaper output by remembering organization-specific workflows and context. The company claims its models can match or outperform frontier labs using up to 100 times fewer tokens, the currency used to run AI queries.

New and more complex AI models are becoming more expensive than previous models, challenging the conventional wisdom that greater scale leads to lower costs.

“We are facing a data explosion and a cost explosion,” said Leigh Marie Braswell, a partner at Kleiner. “Engram comes in and basically maps your organization and delivers much cheaper output.”

Less than a year after its founding, the 13-person company has built a client roster that includes: MicrosoftConcept and legal AI initiative Harvey. Engram, which comes from the neuroscience term that refers to a memory trace in the brain, plans to use the funding to support computing and talent.

Engram co-founder and CEO Dan Biderman has a lifelong obsession with memory. He said it all started when he was a child, trying to trick his amnesiac grandmother into remembering little facts about him and his siblings.

This led Biderman to eventually earn a doctorate in computational neuroscience at Columbia University and later join Stanford University’s artificial intelligence laboratory. Working at Stanford, Biderman began to become aware of what he calls the “genius alien model”: the idea that AI is smart but its memory is much more limited than it seems. At the same time, more context can overwhelm models by requiring more research and reading, as well as higher costs.

Biderman acknowledges that Engram’s models are “definitely not better” than models like OpenAI and Anthropic, but says they excel at specialization, sometimes at the expense of other capabilities.

“We’re trying to go beyond current note-taking and build this layer of intuition that people have that current models don’t,” Biderman said.

WRISTWATCH: The solution to overspending on AI is a problem for OpenAI and Anthropic