Brain for Perplexity Computer

Brain is Perplexity’s newly launched self-improving memory system designed specifically for its multi-model AI agent orchestrator, Perplexity Computer. Released in June 2026, Brain solves a core issue of traditional AI chatbots—forgetting your project context after each chat window closes. Instead of just storing static personal facts (like your name or job title), it centers its memory on the actual work performed by the agent. [1, 2, 3, 4, 5, 6, 7]

• Agent-Centric Work Memory: Brain remembers what methods succeeded, what sources failed, and what corrections you provided. [6, 8]
• The Living Context Graph: Every task executed inside Perplexity Computer links into a visual, continuously expanding graph. [2, 8]
• Overnight Self-Improvement: At set intervals (such as overnight), the system parses the context graph to dynamically upgrade a personal “LLM wiki”. This wiki automatically pre-loads into the agent’s sandbox environment before your next work session. [6, 8]
• Traceable Lineage: In line with Perplexity’s emphasis on transparency, every memory item maps directly back to its source file, data connector, or specific prompt session. [6, 8]

According to first-party benchmarks released in the official Perplexity Blog Announcement, adding Brain to the agent harness yields clear procedural gains: [3, 8, 9]

• +25% Answer Correctness on recurring tasks.
• +16% Data Recall across complex workflows.
• -13% Operational Cost per task due to fewer wasted context tokens and shorter prompts. [8, 10, 11]

Brain is currently in a Research Preview phase. It is available globally for Perplexity Max and Enterprise Max subscribers. You can toggle the memory settings, audit stored context blocks, or delete old entries directly from the “Customize” section located in your sidebar. [2, 7, 10, 12, 13]

I can help you deep dive into utilizing this feature if you let me know:

• What specific workflow (coding, market research, writing) are you building?
• Do you want instructions on connecting external data tools like GitHub, Slack, or Notion?
• Are you looking for help with auditing and deleting unwanted memories?