Article · AI translation
IT Home reported on November 2 that Alibaba Cloud announced today that it recently upgraded its Memory service. For Personal Agent scenarios, the upgrade strengthens its ability to remember user preferences, plans and past agreements, and offers access through MCP, CLI and other methods.
IT Home learned that Memory extracts key information from conversations and saves it as memories and user profiles according to rules. In subsequent sessions, an Agent retrieves relevant memories based on the user's identity and current question and adds them to the task context to generate a response. New, useful information can continue to be written to memory, helping the Agent keep track of user needs and reduce repeated questions.
In recent HaluMem-medium comparison results, Memory received the highest scores on all 4 measures: recall, precision and F1 for memory extraction, and the C metric for memory question answering. In the PersonaMem-v2 evaluation, Memory ranked first among the 8 listed systems with an overall score of 56.82 and received the highest score in 7 of the 9 task dimensions.
Take a financial advisory scenario as an example. When a user says they “prefer a relatively conservative approach, may need the money within 6 months, and do not want large fluctuations,” Memory can save their risk preference, time horizon for using the funds and liquidity needs in a user profile. In a later, new session, the Agent can retrieve the relevant profile and use it alongside product information and business rules to compare products and provide risk reminders. When the user's goals or preferences change, the Agent can confirm the changes before updating the memory. Memory handles storing and retrieving information, while the Agent handles product comparisons and risk reminders.
Developers can customize memory rules and profile fields to specify what information should be saved. They can also view memory contents, debug retrieval results and track changes in the console, making long-term memory manageable and traceable.
For integration, developers can use the API to write, view and semantically search memories. They can also use MCP and CLI to integrate memory capabilities into Agent tools and frameworks such as Codex, DeepSeek Harness, Claude Code and OpenClaw, incorporating them into existing workflows.
Memory can also be used with Skill: Memory stores an individual user's preferences, while Skill packages reusable task workflows. For example, developers can package steps such as reading a user profile, analyzing product risks and checking whether products meet the user's needs into a financial advisory Skill. Each time it runs, it can retrieve the user's latest profile, allowing the service method to be reused across tasks and the service content to be adjusted for each user.