Alibaba Cloud upgrades Memory service to enhance AI agent personalization

Alibaba Cloud has expanded its Memory service, providing developers with tools to equip AI agents with long-term memory and personalization capabilities. The update focuses on the extraction and storage of key information from dialogues, which is then used to populate user profiles for future interactions.

By incorporating these profiles into the task context, agents can tailor their responses without requiring users to repeat information. For instance, in an investment consulting application, an agent can recall a user's stated risk tolerance and liquidity needs to provide relevant, up-to-date guidance. The system includes verification loops to ensure that when user goals evolve, the stored memory is updated accordingly.

From a technical perspective, Alibaba Cloud has introduced support for MCP (Model Context Protocol) and CLI, allowing the memory service to integrate with established frameworks like DeepSeek Harness and Claude Code. Developers retain control over the memory management, with a console provided to debug, trace, and audit the information being retrieved.

The service is designed to complement existing modular task frameworks. According to internal performance data, the Memory service has shown high accuracy rates in memory retrieval and question-answering tasks within standardized benchmarks like PersonaMem-v2.