Memory-Aware Shopping Agents with Strands Agents and Mem0

A research paper from Alibaba proposes a two-stage e-commerce agent that remembers customer preferences across sessions. In this post, you learn how to build it as a working chat app using Strands Agents, Amazon Bedrock, Mem0, and the Shopify Storefront Model Context Protocol (MCP). Why e-commerce agents forget everything Most e-commerce chatbots have goldfish memory. A customer tells your assistant: "I'm a size M, I hate synthetic fabrics, my budget is around $200." Three sessions later, they're back. The bot asks again. This is not only a UX annoyance. It's a conversion problem. And it's entirely avoidable. A paper published in March 2026, Shopping Companion (arXiv:2603.14864) from Alibaba's international commerce team, tackles this directly. The researchers build a large language model (LLM) agent that remembers customer preferences across sessions, retrieves them before searching, and asks the customer to confirm before recommending anything. What the paper proposes The core idea i