AI agents are becoming more capable of carrying out complex tasks on their own. The next challenge is helping them remember what they have already done.
Perplexity has introduced a new system called Brain, a memory layer designed for its AI agent platform, Computer. The company says the feature allows AI agents to learn from previous work, improving their performance over time rather than starting each task from scratch.
According to Perplexity, Brain continuously records the actions taken by Computer, including what worked, what failed, and how tasks were completed. That information is then organized into a growing knowledge graph that the agent can reference when handling future assignments.
The launch marks another step in the industry’s push toward more autonomous AI systems. While most AI assistants today can remember details from conversations, Brain focuses on something different: operational memory. Instead of remembering personal preferences or chat history, it remembers how work gets done.
Perplexity says every task performed through Computer feeds into this memory system. Over time, the platform builds a structured record of completed work, allowing the agent to reuse successful approaches, avoid previous mistakes, and complete similar assignments more efficiently.
The company describes Brain as a continuously learning system that grows alongside the user. In early testing, Perplexity reported improvements in accuracy, recall, and operating costs for workflows that depended heavily on previous task history.
Brain is built on top of Perplexity Computer, the company’s agent platform that combines multiple AI models to perform research, coding, analysis, and other multi-step tasks. Unlike traditional chatbots, Computer is designed to execute actions and complete workflows with limited human involvement.
The announcement comes as AI companies race to solve one of the biggest limitations facing autonomous agents: memory. Many systems can reason through tasks effectively but struggle to retain lessons from previous work sessions. As a result, users often have to repeat instructions or rebuild context every time a new task begins.
Perplexity believes Brain can help address that problem by turning past activity into a reusable knowledge base. The approach reflects a broader industry trend as AI companies look beyond chat interfaces and build systems capable of acting, adapting, and improving through experience.
Brain is currently available in research preview for Perplexity Max and Enterprise Max subscribers.
The bigger ambition is clear. As AI agents take on more work traditionally handled by humans, memory may become just as important as intelligence itself.
Read also: Malta proposes treating DeFi decentralisation as spectrum under MiCA framework