Perplexity AI is rolling out a major upgrade to its Comet Assistant and core answer engine,introducing a new memory layer that can remember user preferences, interests, and past conversations to deliver more personalized, context-aware responses. The feature is designed to make AI feel less like a generic chatbot and more like a consistent “second brain” that evolves with each interaction.​

Perplexity unveils AI with built‑in memory

Perplexity’s new personalization system continuously captures key details from user interactions—such as favorite brands, dietary requirements, recurring topics, and long‑running projects—and turns them into a structured memory store. Instead of requiring users to re-explain themselves or manually summarize previous chats, the assistant automatically pulls in this context to keep conversations flowing and reduce repetition.​

This approach targets a common limitation of traditional large language models, which often hit context limits mid-workflow and force users to start over or engineer prompts just to keep the model on track. By pre-loading the most relevant details in the background, Perplexity aims to preserve continuity and keep users in a productive “flow state” across sessions.​

How the new memory layer works

Unlike many AI products that treat user history as just another slice of training data, Perplexity separates memory into its own retrieval layer. When a user asks a question, the system fetches only the most relevant stored preferences and past exchanges, then uses them directly to shape the answer, rather than relying purely on statistical patterns.​

Practical use cases for everyday tasks

Perplexity positions the memory upgrade as a step toward an AI that can participate in ongoing projects rather than answering isolated questions. A user planning a trip can ask for a book recommendation for an upcoming flight, then later request restaurant ideas or packing tips that reflect earlier preferences like budget, dietary needs, or travel style.​

Privacy controls and user choice

Perplexity emphasizes that users remain in control of what the assistant remembers and how that data is used. Memory can be switched off entirely, individual memories can be managed or removed in account settings, and AI data retention for model improvement can be disabled via dedicated privacy controls.​

Context that travels across models

One of the more distinctive aspects of Perplexity’s rollout is that the memory layer works across every model available in the product, including different reasoning, fast, and specialized models. Users can switch models from question to question—choosing, for instance, a more powerful model for complex research and a lighter one for quick lookups—without losing their accumulated context.​

This “context portability” means that the hours spent building up a personal knowledge base with the assistant carry forward as new models launch or new capabilities are added, including within the Comet AI browser agent. Perplexity frames this as both a productivity benefit and a way to protect users’ previous work, turning scattered interactions into a persistent, coherent intelligence layer that grows alongside the way each person thinks and works.​