Perplexity rolls out a clear, practical update: its AI assistants, like Comet Assistant, now remember preferences and conversations to deliver faster, more personalized answers. Published on November 26, 2025, the change aims to solve the everyday problem of losing the thread when you work with models that forget what you already said.
What it means for an assistant to have memory
Have you ever felt like every chat with an AI is starting from scratch? Memory exists to avoid exactly that. Instead of relying only on what you type each session, the assistant stores useful details—like your favorite brands, dietary restrictions, or ongoing projects—and uses them to give more relevant answers without making you repeat everything.
It’s not magic: it’s continuity. Memory acts like pre-loaded context that keeps your workflow moving so you don’t have to write manual summaries or repeat instructions.
How Perplexity implements memory
Perplexity says it doesn’t treat your history as if it were massive training data. Rather than folding everything into a dataset to recompute probabilities, the system retrieves relevant information from a memory store and uses it directly to build the response.
That changes the nature of the output: it’s not just a statistical guess about what most would say, but a reply informed by your specific history. In practice that means answers that are more accurate and in a tone that better matches how you like to work.
Memory turns scattered context into coherent intelligence, and the assistant evolves with you.
Practical examples
Recommend running shoes.- The assistant considers whether you’re injured, training for a marathon, or prefer casual runs.I’m flying to NYC next week. Suggest a book for the flight.- Suggestions based on your past tastes.What gifts should I buy for my mom and my brother?- Ideas aligned with preferences you’ve already shared.What advice did you give last week about that difficult coworker?- It retrieves earlier advice and summarizes or updates it.
These examples show how memory turns repetitive interactions into continuous, useful conversations.
Privacy and control
Sound like the AI will remember everything without asking? Perplexity provides clear controls: you can choose what to remember, turn memory off whenever you want, and use an incognito mode where both memory and history are disabled automatically.
Data is encrypted, and there’s an option to not contribute to improving models via the AI Data Retention setting. In short: memory yes, but with tools to keep you in control.
Portability of context between models
A practical detail: memory isn’t locked to a single model. Perplexity lets the stored context be used with different models depending on the task: reasoning models for complex analysis, fast models for quick queries, or specialized models for specific domains.
That protects your work: the hours you spent building a profile or memory remain valuable even if you switch models or a new one arrives.
Why it matters today
The difference isn’t just technical, it’s experiential. If you work on long projects, manage recurring tasks, or want an assistant that remembers personal nuances, memory boosts efficiency and relevance. Who wouldn’t prefer answers that already know a bit about you?
The arrival of memory in assistants like Comet Assistant shows AI moving from a one-off tool toward becoming a work partner with continuity.
Original source
https://www.perplexity.ai/hub/blog/introducing-ai-assistants-with-memory