Alibaba Cloud Memory Service Upgrade: Enhancing Long-Term Memory and Personalization for Personal Agents
Alibaba Cloud announced an upgrade to its Memory service, enhancing the ability to remember user preferences, plans, and historical agreements for Personal Agent scenarios, and providing access methods such as MCP and CLI. Memory extracts key information from conversations and saves it as memories and user profiles according to rules; in subsequent sessions, it retrieves information based on user identity and the current question and adds it to the task context. In the HaluMem-medium comparison, Memory achieved the highest scores in all four metrics for memory extraction: recall, accuracy, F1, and the memory question-answering C metric; in the PersonaMem-v2 evaluation, it ranked first among the 8 systems listed with an overall score of 56.82.
Why read this
AI score 55
The original article provides the specific mechanisms for memory extraction and profile retrieval, along with two evaluation comparisons, which can be used to assess how long-term memory is implemented in Personal Agents.
Topics
Body · AI translation
Alibaba Cloud Memory service upgraded, enhancing long-term memory and personalization capabilities for Personal Agents
English text is being prepared.
English text is being prepared.
English text is being prepared.
In recent HaluMem-medium comparison results, Memory achieved the highest scores in all four metrics: recall, accuracy, and F1 for memory extraction, as well as the memory question-answering C metric; in the PersonaMem-v2 evaluation, Memory ranked first among the 8 systems listed with an overall score of 56.82, and achieved the highest scores in 7 of the 9 task dimensions.
Taking the financial advisory scenario as an example, when a user states "prefers stability, may need the money within half a year, and does not want too much fluctuation," Memory can save the risk preference, fund usage horizon, and liquidity needs as a user profile. In a later new session, the Agent can invoke the relevant profile and, combined with product materials and business rules, carry out comparisons and risk reminders. When the user's goals or preferences change, the Agent can first confirm and then update the memory. Among this, Memory is responsible for information storage and retrieval, while product comparison and risk reminders are completed by the Agent.
Developers can customize memory rules and profile fields, specify the information that needs to be saved, and view memory content, debug retrieval effects, and track changes in the console, making long-term memory manageable and traceable.
In terms of access, developers can complete memory writing, viewing, and semantic retrieval through APIs, and can also integrate memory capabilities into Agent tools and frameworks such as Codex, DeepSeek Harness, Claude Code, and OpenClaw through MCP and CLI, integrating them into existing workflows.
Memory can also be used in conjunction with Skill: Memory saves the user's individual preferences, and Skill encapsulates reusable task processes. For example, developers can encapsulate steps such as reading user profiles, analyzing product risks, and checking requirement matching into a financial advisory Skill, invoking the user's latest profile each time it is executed, so that service methods are reused with tasks and service content is adjusted with the user.