Sarah Wooders

Sarah Wooders is co-founder and chief technology officer of Letta, a company that builds AI agents with long-term memory. It grew out of MemGPT, an open-source project she created with Charles Packer at UC Berkeley's Sky Computing Lab. Her own site describes the goal as "building agents that learn over time, through memory that is model-agnostic and interpretable."

Tagging Product Data at Glisten

Wooders studied computer science and mathematics at MIT. After graduating she co-founded Glisten AI with Alice Deng, and the company went through Y Combinator's Winter 2020 batch. Glisten used computer vision and natural language processing to break product photos and descriptions into structured attributes for e-commerce. The founders learned that their buyers were further from the shopper than expected. "What we realized over time was that the right customer is the customer who feels the pain point of having messy unreliable product data," Wooders told TechCrunch in 2020, noting that the first customer was a pricing optimization company.

Moving Data Between Clouds at UC Berkeley

She then began a PhD at UC Berkeley, advised by Ion Stoica and Joseph Gonzalez in the Sky Computing Lab, whose earlier incarnations produced Ray and Spark. With fellow student Paras Jain she co-created Skyplane, an open-source tool that moves large amounts of data between cloud providers and regions. In her applied machine learning work she had often switched between AWS and Google Cloud, she told Decibel, and the biggest problem she saw was "data gravity," the difficulty of moving data from one provider to another. The project reported transfers up to 110 times faster than AWS DataSync.

Building MemGPT With Charles Packer

At Berkeley, Wooders and Packer created MemGPT, which manages data and memory so that AI agents and chatbots can remember earlier users and conversations, a gap in language models that on their own are stateless. The lab around them had already produced vLLM for serving models, and the two began to question that focus. "Why are developers interfacing with LLMs? It makes a lot more sense if developers are instead interfacing with agents as services as opposed to just LLMs as services," she said in a Felicis conversation with Stoica and Packer. She credits Stoica with a lesson she carried into the company: "it's very important to choose the right problem to work on. And that applies both for research and startups."

Launching Letta

Letta came out of stealth in September 2024 as the commercial company behind MemGPT, with a $10 million seed round led by Felicis, TechCrunch reported. Moving from research to a startup changed her days most of all: in grad school "you spend a lot of time on research, getting to read papers," she said, while a startup brings "a lot more time on just operational things like recruiting, marketing." Her open-source work now includes Letta Code, which she describes as a memory-first coding agent.

Asked in the Felicis conversation to pick a research paper that excited her, she chose the sleep-time compute paper. She expects agent capabilities to keep improving regardless of the models underneath. "Even if improvements in models stall, I think we will just get better and better at using the models," she said, adding that this is "a big thing that we work on at Letta."