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

In a rapidly evolving landscape where corporate America is tightening its grip on uncontrolled AI development, a new player named Engram is seizing the moment. Founded just eight months ago, this startup has secured a remarkable $98 million in funding from notable investors such as General Catalyst, Kleiner Perkins, and Sequoia, along with support from OpenAI co-founder Andrej Karpathy, who has recently joined Anthropic. Engram positions itself as the 'learned memory' of AI, introducing innovative models that can remember specific organizational workflows and context. This capability allows them to predict questions and deliver more intelligent responses using significantly fewer resources. The startup claims its technology can either match or even surpass the performance of leading research labs while utilizing up to 100 times fewer tokens, the essential currency for executing AI queries. The surge in new AI models has brought escalating costs, contradicting the traditional belief that scaling would lead to reduced expenses. As Leigh Marie Braswell, a partner at Kleiner Perkins, points out, "There's an explosion of data and cost," emphasizing that Engram’s solution offers a more economical alternative by effectively mapping out organizational needs. Despite being less than a year old, Engram has already attracted a diverse clientele, including industry giants like Microsoft, Notion, and legal AI innovator Harvey. The funding will primarily be directed toward enhancing their computational capabilities and expanding their team. Co-founder and CEO Dan Biderman's passion for memory traces back to his childhood, where he sought ways to help his grandmother, who struggled with memory loss, recall details about their family. This early fascination led him to earn a PhD in computational neuroscience from Columbia University and to contribute to Stanford University's AI lab. During his time there, he developed the concept of the 'genius stranger model,' which highlights the intelligence of AI while acknowledging its limited memory capacity. Biderman candidly admits that while Engram's models might not outclass those from industry leaders like OpenAI and Anthropic, they do specialize in certain areas, enhancing their effectiveness in specific contexts. "We aim to transcend conventional note-taking by creating a layer of intuition akin to human memory, which current models lack," he stated, emphasizing the startup's innovative vision.