Perplexity Introduces "Brain" - A Self-Improving Memory System for AI Agents

Perplexity has announced Brain, a new memory system that helps AI agents learn from their work and continuously improve over time.

Unlike traditional AI memory, which mainly remembers user preferences and personal information, Brain focuses on remembering:

What tasks were completed
What worked and what failed
User corrections and feedback
The best sources, tools, and workflows

How it works

Brain creates a "context graph" of the work performed by the AI agent. During scheduled reviews (such as overnight), it analyzes past tasks and updates its knowledge, allowing future tasks to be completed faster and more accurately.

Early Results

25% improvement in answer correctness on familiar tasks
16% better recall of relevant information
13% reduction in costs for tasks requiring historical context

Why this matters:

This is a step toward truly proactive AI systems that don't just respond to prompts but continuously learn from experience, improve workflows, and become more effective over time.

The more you use the system, the smarter and more efficient it becomes.

Brain is currently rolling out in Research Preview for Perplexity Max and Enterprise Max users.

What are your thoughts on self-improving AI memory? Could this be the next major leap in AI agents?

1 comment

Perplexity Introduces "Brain" - A Self-Improving Memory System for AI Agents