Garry Tan, President and CEO of Y Combinator, recently announced a significant intelligence upgrade to his personal AI agent setup, powered by OpenClaw and Hermes Agent frameworks. Tan stated on social media, > "My OpenClaw/Hermes Agent setup just got way smarter," signaling enhanced capabilities for his sophisticated autonomous systems. This improvement is largely attributed to updates within GBrain, his open-source knowledge layer designed for persistent AI agent memory.
GBrain functions as a production-grade AI memory system, transforming extensive Markdown notes into a searchable, reasoning-capable, and continuously evolving knowledge graph. It serves as the long-term memory for AI agent platforms like OpenClaw and Hermes Agent, providing them with accumulated experience and context. This system is not merely a demonstration but represents Tan's actual daily-use infrastructure, managing vast amounts of his personal and professional data.
Recent updates, specifically GBrain v0.12, have introduced a self-wiring knowledge graph, yielding measurable performance enhancements. The update boasts a 5% improvement in precision, an 11% increase in recall, and a notable 28% better graph search capability. These metrics suggest a more efficient and accurate retrieval of information, directly contributing to the agents' improved intelligence.
The setup integrates with both OpenClaw and Hermes Agent, two prominent open-source AI agent frameworks that approach autonomous tasks differently. OpenClaw is often characterized by its strict instruction-following, while Hermes Agent emphasizes a learning loop that creates and refines skills through experience. Tan has previously expressed a preference for the "free" nature of OpenClaw/Hermes Agent over more restrictive alternatives.
This advancement underscores the ongoing development in personal AI agents and open-source contributions to the field. By enhancing the core memory and reasoning capabilities of these agents through GBrain, Tan's work highlights a path toward more autonomous and intelligent systems. The project, released under an MIT license, allows others to leverage this sophisticated infrastructure for their own AI agent deployments.
