Perplexity introduces AI assistants with memory for more personalised answers
Perplexity has launched AI assistants with built-in memory, offering more personalised and context-aware answers for users.
Perplexity has unveiled a memory-driven personalisation layer for its AI assistants that remembers user preferences, interests, and relevant details from past conversations to deliver faster, more tailored answers.
The feature forms part of the company’s broader push into agentic AI through its Comet Assistant.
The company said its assistants now synthesise key context automatically—such as favourite brands, dietary needs, and recurring keywords—and pre‑load this when tackling new queries.
Rather than treating chat history like generic training data, Perplexity retrieves relevant context from a user’s own memory store to improve precision and continuity across sessions.
Perplexity, led by co‑founder and CEO Aravind Srinivas, positioned this update as a step toward assistants that operate as a “second brain” and operator for everyday tasks and research.
How it works across models
Memory is described as model‑agnostic: users can switch between reasoning‑heavy models, faster models, or specialised models while retaining the same personal context.
Perplexity framed this as “context portability”, allowing work to carry over as models evolve.
How it works:
- Pre‑loading context: The assistant pre‑loads the most pertinent details from prior chats so users need not repeat themselves when resuming a thread.
- Targeted retrieval: Instead of probabilistic guesses from historical data, the system retrieves specific items from a private memory store to shape responses.
- Example prompts: Product recommendations that reflect training status or past injuries, trip reading suggestions tuned to taste, and holiday gift ideas informed by prior exchanges.
Privacy, control, and transparency
Perplexity has emphasised user control. Memories can be turned off, while incognito mode automatically disables both memory and search history.
The company also said data is encrypted and users may opt out of contributing to model improvement via the AI Data Retention setting.
A Help Centre article additionally explains that memories and prior searches may be referenced to personalise answers, and can be viewed and cleared in Settings.
Perplexity's launch comes amid rivals also expanding long‑term memory features this year.
Anthropic has rolled out memory for Claude to Team and Enterprise users and, more recently, to Pro and Max plans, with granular controls and incognito chats.
Microsoft also introduced long‑term memory and refreshed controls in Copilot, alongside documentation explaining storage and admin policies.
Additionally, OpenAI updated ChatGPT with project‑scoped memory options and a daily Pulse preview that draws on memory and chat history.