Perplexity has launched Brain, a memory system for its Computer agent that builds a context graph of past sessions, sources, and corrections, synthesizing it overnight into a personal LLM wiki. This wiki is loaded before each new task, providing full context. Early metrics show a 25% boost in answer correctness, 16% recall improvement, and 13% cost reduction on context-heavy tasks. Brain is available for Max ($200/month) and Enterprise Max subscribers in Research Preview. Unlike self-hosted solutions like OpenClaw (379k GitHub stars) and Hermes, Brain runs entirely on Perplexity's cloud, offering less data sovereignty. Brain improves performance on repeated tasks but does not enhance cross-domain generalization. The system aims to make the agent more efficient by remembering work-related context rather than user preferences.

Key facts

• Brain builds a context graph of sessions, sources, and corrections, synthesized into a personal LLM wiki.
• Metrics: 25% higher answer correctness, 16% recall improvement, 13% cost reduction on repeated tasks.
• Available now in Research Preview for Max and Enterprise Max subscribers ($200/month).
• Unlike self-hosted tools (OpenClaw, Hermes), Brain runs on Perplexity's cloud with less user control.
• Brain improves performance on repeated tasks but does not enable cross-domain generalization.