Perplexity Launches Brain Memory System for Computer to Retain Context Across Tasks.

Perplexity has released Brain, a continuously learning memory system that gives its Computer agent persistent context across tasks. The feature, announced in a LinkedIn post, is available now as a research preview for Perplexity Max subscribers.

Brain builds a context graph from user interactions, allowing Computer to start each task with full awareness of previous projects, decisions, and sources instead of working from scratch. According to Perplexity, this persistent memory improves answer correctness by 25%, boosts recall by 16%, and reduces cost per task by 13% on queries requiring past context.

Every task run through Computer now feeds into Brain's context graph. When you ask Computer to generate code, research a topic, or analyze data, Brain stores relevant information and connections. The next time you invoke a related task, Computer pulls from this accumulated knowledge.

Memories are transparent and traceable. Each stored piece of information links back to its origin, whether that's a specific session, uploaded file, or external source. Users can review and manage saved memories through the "Customize" section in the sidebar, giving them control over what Brain retains.

The architecture marks a shift from stateless AI assistants to agents that build understanding over time. Rather than treating each query in isolation, Brain lets Computer recognize patterns, remember user preferences, and maintain continuity across work sessions.

Brain is currently limited to Max subscribers and carries the "research preview" label. This suggests Perplexity is still refining the system's behavior, privacy controls, and scalability before a wider rollout.

The performance metrics, while promising, apply specifically to "tasks that require past context." Perplexity has not disclosed how Brain performs on standalone queries or whether the system introduces latency for users who prefer ephemeral interactions.

Memory systems in AI tools raise questions about data retention, cross-session privacy, and unintended bias from accumulated context. Perplexity's decision to make memories inspectable and removable addresses some of these concerns, but the long-term implications of persistent AI memory remain an open area.

Brain is the latest addition to Perplexity's Computer product, which recently integrated Deep Research and added Claude Fable 5 as an orchestrator model. These updates position Computer as more than a search interface, moving it toward a stateful agent that learns user workflows and adapts over time.

For Max subscribers working on multi-step projects, research tasks, or iterative development, Brain could reduce repetitive context-setting and improve output relevance. Whether it justifies the subscription cost will depend on how well the memory system integrates with individual workflows and whether users trust the system to manage their context accurately.

The feature is live now in the Perplexity interface under the "Customize" sidebar option.