Perplexity details Brain, the self-improving memory system in Perplexity Computer, and Agent API memory tests

Perplexity on November 2, 2026, published a technical guide to persistent AI agent memory. The guide covers the Brain system in Perplexity Computer, the memory-building components of the Agent API, and tests for whether stored context improves an agent's later work. Unite.ai reported on the publication.

What Perplexity describes

Perplexity describes Brain as its self-improving memory system for Computer. According to the company, Brain is built to organize past work and corrections into an evolving knowledge wiki. Perplexity says it uses token-level memory.

The guide also covers the Agent API, specifically the components it provides for building memory. Perplexity pairs this with a testing approach: measuring whether context an agent has stored actually improves how it performs on later tasks.

Details not specified

Pricing, context window size, benchmark scores and availability terms for the memory features are not yet disclosed in the information available at publication time. No new model accompanies the guide.

What this means

The guide treats agent memory as something to be tested, not assumed. Perplexity frames the question as whether stored context improves later work, which gives developers building on the Agent API a criterion for evaluating their own memory setups. Brain's design, as Perplexity describes it, organizes both past work and corrections into a knowledge wiki, so user corrections are part of what the system retains. Teams assessing the approach should look at Perplexity's full guide for the architecture and test results.