A startup named Letta has emerged from stealth mode, unveiling technology that enables AI models to remember users and past conversations. Developed in UC Berkeley’s renowned startup incubator, Letta secured $10 million in seed funding led by Felicis’ Astasia Myers, boasting a $70 million post-money valuation. The startup has garnered support from notable angel investors in the AI space, including Google’s Jeff Dean and Hugging Face’s Clem Delangue. Founded by Berkeley PhD students Sarah Wooders and Charles Packer, Letta is a product of the Sky Computing Lab and the commercial offshoot of the widely popular MemGPT open source project. The Sky Computing Lab, led by professor Ion Stoica, has a history of producing successful companies and advanced open-source language model projects. MemGPT aims to overcome the stateless nature of conventional large language models (LLMs), like ChatGPT, which struggle to remember user data over time. This capability is crucial for applications like customer support and healthcare symptom-tracking. The interest in MemGPT was so intense that its whitepaper went viral on Hacker News before the official release. After gaining significant attention, the founders quickly prepared the launch of MemGPT on GitHub, which garnered thousands of stars and forks, alongside interviews and tutorials.

Myers, who discovered the project through the whitepaper, recognized its commercial potential and reached out to the founders. The trio conducted meetings with various VCs before choosing Felicis, supported by a network of influential investors facilitated by Stoica. Currently, Letta is preparing for the launch of its commercial product, Letta Cloud, which will offer a service for developers to deploy stateful AI agents via REST APIs and manage long-term data. While MemGPT is already in use, Letta Cloud will open for beta users shortly. Wooders highlights the primary application of MemGPT as creating engaging, personalized chatbots, including specialized functions such as assisting cancer patients by learning from their medical history. However, competition is fierce, with companies like LangChain and OpenAI’s Assistants API already offering similar services. Despite the competitive landscape, Letta distinguishes itself by claiming compatibility with any AI model, unlike OpenAI, which is confined to its AI offerings. Moreover, Letta is committed to open-source principles, advocating for transparency over proprietary solutions, which co-founder Packer believes is essential for effective AI application development.