Letta: Memory Layer

Letta is an innovative AI startup founded by Berkeley PhD students Sarah Wooders and Charles Packer, emerging from stealth in September 2024 with USD 10 million in seed funding led by Felicis. The company originated from the MemGPT research project at UC Berkeley's AI Research Lab and focuses on building stateful AI agents with advanced memory systems that can learn and self-improve over time. Letta's platform enables AI agents to maintain sophisticated memory structures—including entities, facts, and timelines—that can be queried and controlled, providing extreme interpretability with learned information stored as human-readable text.

Letta has achieved a post-money valuation of USD 70 million, demonstrating strong investor confidence in its approach to solving one of AI's fundamental challenges: giving agents the ability to remember and learn from past interactions. The company's technology supports advanced, structured memory models that developers can directly inspect, evaluate using LLM-as-judge techniques, or manually review. This transparency and control make Letta particularly valuable for building production AI systems where understanding agent behavior is critical.

Unlike plug-and-play memory solutions, Letta provides a strong architectural backbone that balances open tooling with production pragmatism. The platform shines for stateful AI agents requiring robust, developer-friendly memory management, though developers still need to make architectural decisions around vector stores, RAG strategy, and observability. Letta's approach represents a middle path—more structured and extensible than simple memory layers, but less complex than research-heavy stacks, making it ideal for teams building sophisticated AI agents that need to maintain context over extended periods.