Self-Improving Memory for Agents
Perplexity, Thursday, June 18th, 2026
Perplexity unveils Brain, a self-improving memory system that helps AI agents learn from their work over time.
Perplexity introduced Brain, a new approach to memory in AI built for its Computer agent product. Brain constructs a context graph of the work an agent performs and periodically reviews it, such as overnight, to teach itself how to do the work better.
It remembers what the agent did, what worked, what failed, and what corrections were made, enabling continuous improvement. The system aims to make agentic workflows more reliable and capable as they accumulate experience.