Letta, a UC Berkeley spin-off, emerged from stealth with $10M in seed funding

Letta (an approximation of the word 'letter') recently emerged from stealth with $10 million from Felicis, with the support from Sunflower Capital, Essence VC, and angels including Jeff Dean (Google DeepMind), Clem Delangue (HuggingFace), Cristobal Valenzuela (Runway), and others. The much-anticipated startup is led by the creators of the popular open-source project MemGPT; Letta will continue the research started with MemGPT and incorporate the advances into a production-ready platform that lets developers go beyond stateless API calls by allowing the creation and deployment of stateful agents and LLM APIs.

OpenAI's o1 launch brought undisputable evidence of the turn from scaling at training time to scaling at inference time. We can easily expect more developments related to processes like extended chains of thought, reflection, and other methods that provide more resources to the models during inference to improve their reasoning capabilities. Letta's mission aligns with this trend, but it also has set its sights on more ambitious goals: a latent danger with having closed-source model providers dominate the playing field means that we risk having any outputs related to these techniques, including memory, hidden by providers, essentially locking customers into their platform.

To address this, Letta looks to establish itself as a model-agnostic, open-source alternative that provides a transparent, flexible, and controllable way to build applications incorporating memory and state while retaining the freedom to swap models without information loss. Letta currently offers its API platform (powered by MemGPT) free of charge and is onboarding developers for early access to its Letta Cloud platform and Agent Development Environment.