Perplexity's AI Now Remembers Its Fails — Finally, an Agent With Better Notes Than Its Users 🧠

Perplexity launched Brain on Thursday, a memory system for its Computer agent that retains context across sessions by logging task outcomes, source reliability, and user corrections rather than personal user data. "With Brain, Computer starts each task with full context of your projects, decisions, and sources instead of from scratch," Perplexity said. "Every memory links back to the session, file, or source it came from with full transparency and control."

The system builds a context graph tracking which connectors were used, which sources held up, what corrections users made, and which approaches failed. At set intervals, overnight by default, Brain synthesizes the graph and updates a personal LLM wiki that loads into Computer's sandbox before the next task begins. Each memory entry links back to the originating session or file, allowing users to trace any decision to its source.

Perplexity's internal metrics show Brain improves answer correctness by 25% on tasks Computer has already handled, boosts recall by 16%, and reduces the cost of context-heavy tasks by 13%. The company described the figures as internal numbers rather than third-party benchmarks.

Brain enters a market that already includes self-hosted alternatives. OpenClaw, which has accumulated over 379,000 GitHub stars since launch, has used markdown files and a SQLite database with FTS5 full-text search to persist context across sessions, adding "providence labels" in April 2026 that tag each stored memory as observed, user-confirmed, model-inferred, or imported from a transcript. Nous Research's Hermes, released in February 2026, extracts reusable reasoning patterns after each completed task and writes them as skill files in plain markdown, loading relevant skills on subsequent tasks instead of reasoning from scratch.

Unlike those open-source tools, which run on user hardware, Brain is operated by Perplexity, a multibillion-dollar company, placing it in the category of managed agent services rather than self-hosted infrastructure.