Perplexity has launched a novel memory system called Brain, which is designed to enhance the efficiency of its AI agent, Computer. Unlike traditional AI memory systems that focus on user preferences, Brain concentrates on the actions and outcomes of the agent's work. This self-improving memory constructs a context graph that tracks the agent's activities and learns from them overnight. By reviewing this graph at regular intervals, Brain aims to refine the agent's performance over time. This system is currently available to Perplexity Max and Enterprise Max subscribers in Research Preview.
Brain operates on two main axes: the subject of memory and its purpose. While traditional AI memory centers around the user, storing their preferences and roles, Brain focuses on the agent's tasks, remembering what strategies were effective or not. This shift in focus aims to enhance the agent's performance rather than just user engagement.
The context graph created by Brain allows Computer to navigate the user's environment and learn from past experiences. Each memory entry is traceable back to its source, which aids in debugging and builds trust. As users interact with Computer, Brain continuously improves, learning from both successes and failures, ultimately leading to more efficient task execution.