Perplexity Adds Self-Improving Memory 'Brain' to Computer, Boosting Task Correctness by 25%
On June 18, 2026, Perplexity announced Brain, a new memory system for its Computer agent platform. Unlike typical “memory” features that recall user preferences, Brain learns from the agent’s own work — tracking what succeeded, what failed, and what corrections were made — and improves autonomously over time.
Details
- Context graph: Combines session history, connector results, document changes, and user corrections into a continuously updated context graph, functioning like an “LLM wiki” of the user’s projects, ideas, people, and information sources
- Overnight self-improvement cycle: At set intervals, such as overnight, Brain reviews the graph, analyzes what worked and what didn’t, and refines how it approaches future tasks
- What it learns from: Learns from user corrections and dead-end sources alike, reducing unnecessary model calls and token usage
- Measured results: 25% increase in answer correctness on previously-seen tasks, a 16% improvement in recall, and a 13% cost reduction for tasks requiring historical context
- Availability: Rolling out as a Research Preview for Perplexity Max and Enterprise Max subscribers
How to try it
- Currently available as a Research Preview for Perplexity Max and Enterprise Max subscribers
- Requires use of Computer or Personal Computer — no additional setup is needed to benefit from it
- Designed to keep improving with continued use, so long-term users are likely to see the biggest gains