Perplexity Brain: Self-Improving Agent Memory Systems Are Here

Perplexity has introduced Brain — a self-improving memory system for AI agents that learns from every interaction and gets better over time.

HOW IT WORKS

Brain builds a context graph of everything the agent does on your computer. At set intervals — such as overnight — Brain reviews the context graph and teaches itself how to do the work better. Agents learn which sources lead to the best outputs, remember when a user made a correction, and avoid dead ends that wasted turns in previous sessions.

The result: fewer turns, fewer model calls, better outputs, and lower costs. The feedback loop is continuous — the more work you do, the more efficient Brain makes your agent.

THE HERMES PARALLEL

The concept will sound familiar to anyone following the agent ecosystem. Hermes Agent has been doing something similar — learning from past sessions, remembering mistakes, and continuously improving — but running locally. Perplexity Brain does this in the cloud.

Access requires a Mac or Enterprise plan (higher-tier subscriptions).

KEY TAKEAWAY
Self-improving memory is becoming table stakes for AI agents. The agents that will win are not the ones with the best single-turn performance — they are the ones that learn from every interaction and get measurably better over time. Brain is Perplexity's bet on that thesis.