Perplexity Launches Brain, a Self-Improving Memory System That Builds a Context Graph of an Agent's Work and Learns Overnight - MarkTechPost

Perplexity has unveiled Brain, an innovative self-improving memory system specifically designed for its AI agent, Computer. Unlike traditional AI memory systems that prioritize user information and preferences, Brain is engineered to learn from the agent's own operational history and performance. It achieves this by creating a "context graph" that meticulously records the agent's activities, including successful actions, encountered failures, and any user-initiated corrections. This accumulated data is then processed periodically, often overnight, to synthesize new insights. These insights are subsequently used to update the agent's internal knowledge base, allowing Computer to approach future tasks with a rich understanding of its past experiences. This process aims to eliminate the need for the agent to start from scratch with each new task, leading to improved efficiency and accuracy. Early internal testing conducted by Perplexity has indicated significant performance improvements. These include a reported 25% increase in answer correctness for repetitive tasks, a 16% enhancement in recall capabilities, and a 13% reduction in costs for workflows that demand extensive historical context. Brain is currently being rolled out in Research Preview and is accessible to subscribers of Perplexity Max and Enterprise Max, marking a notable advancement in AI agent learning and adaptive performance.