Perplexity is rolling out a new memory feature for its Computer agent called Brain, a system the company says is designed to help the software learn from past work and improve over time. The feature is available now in Research Preview for Max and Enterprise Max subscribers.
Perplexity describes Brain as a self-improving memory layer for Computer, its AI product for carrying out tasks. Rather than focusing primarily on remembering user preferences, the system is built to remember what the agent did, what succeeded, and where it went wrong. The company says that allows Computer to build on prior sessions and become more effective with continued use.
The system works by creating a context graph of the work Computer performs. At intervals such as overnight, Brain reviews that graph and uses it to refine how the agent handles future tasks. Perplexity says the result is a better starting point for each new request, with the system able to move faster toward answers, use more reliable sources, and avoid wasting time and tokens.
Perplexity argues that most AI memory systems are centered on the user, storing details such as tastes, working style, contacts, and role. Brain instead focuses on the agent's own activity. According to the company, that gives the memory system a different purpose. It is not just meant to make the AI feel more personalized. It is meant to help the agent get better at doing the work itself.
The company says Brain updates its knowledge using signals from sessions, connector results, source document changes, and corrections made by users. Those updates feed into a living context graph that Perplexity describes as a kind of automatically maintained LLM wiki loaded into the agent sandbox.
That wiki includes pages tied to ideas, people, projects, and other parts of a user's work world. Perplexity says the system can use that structure to navigate the user's context and understand what information is most relevant for a given task.
Perplexity says early internal measurements show Brain improved answer correctness by 25% on tasks Computer had seen before. The company also reported a 16% increase in recall and a 13% reduction in the cost of tasks that depend on historical context.
The company says those gains become stronger for users over time as Brain learns more about their work patterns and the surrounding context. Perplexity also says each memory entry links back to the session, file, or source it came from, allowing users to trace where the information originated.
Perplexity framed Brain as a step toward more proactive AI systems that can identify opportunities or problems without being explicitly asked. The company says continuous learning is necessary for Computer to become the kind of agent users and businesses want, especially in organizational settings where the software can be trained on local workflows and sources.
Brain is the latest addition to Perplexity's Computer product, and the company says more capabilities are planned. For now, the feature is limited to Max and Enterprise Max subscribers in Research Preview as Perplexity tests how the memory system performs in real use.