Perplexity Brain Boosts AI Correctness 9.3 Points in First Self Improving Memory System
Products36d agoThe company's new memory system compiles user sessions, files, and sources into a structured knowledge wiki to increase the accuracy of its AI agents. Perplexity reported that Brain improves correctness by 9.3 points, recall by 8.9 points, and currentness by 8.0 points while using 15% fewer tokens. The system is provided to Max users as a way to maintain factual consistency and reduce the cost of processing long user histories.
The architecture utilizes a filesystem of linked Markdown files to build multi-hop context graphs. Background "Dream" agents run offline with specific guardrails to synthesize new information and update the wiki in a continuous loop, allowing the AI to explore history on demand and verify claims against original evidence without loading full datasets into every prompt.
Key sources
- SOURCE@perplexity_ai“improving correctness by 9.3 points, currentness by 8.0, and recall by 8.9 with 15% fewer tokens”x.com
- SUPPORT@perplexity_ai“durable memory storage, foreground agents that answer queries, and background agents that update and improve memory”x.com
- SUPPORT@perplexity_ai“stores memory as a filesystem of linked Markdown files”x.com
- SUPPORT@perplexity_ai“Dream agents run offline, with carefully defined scope and Dream-specific guardrails”x.com
- SUPPORT@perplexity_ai“without loading all of it into every prompt”x.com
- SOURCE@aravsrinivas“builds multi-hop context graphs in a perpetual compounding loop”x.com