On April 9, the Alibaba Cloud BaiLian platform officially launched the "Memory Library" feature, aiming to solve the problem of AI agents forgetting in multi-turn conversations, enabling them to have long-term memory across sessions. This feature is currently available for free to users on a limited basis. Developers can directly call it via API, or deploy it with one click in products like OpenClaw.

The BaiLian Memory Library system includes four core modules: "extraction, storage, retrieval, and injection." After a conversation ends, the system automatically extracts key information based on preset rules and stores it. When a user issues a new instruction, the system precisely retrieves relevant memories through semantic retrieval and injects them into the context in real time, thereby achieving highly personalized responses. To lower the development barrier, this feature supports white-box configuration, allowing developers to customize memory rules based on dimensions such as user personality, consumption habits, and specific family relationships. In addition, the memory library includes general rules covering mainstream scenarios such as consumer electronics, customer service, sales, and emotional companionship, effectively reducing configuration costs by 50%.

In terms of performance, Alibaba Cloud has optimized the search algorithm specifically. Data shows that the average response time (RT) for memory search has decreased by 50%, date relevance has increased by 66%, and memory content relevance has improved by 39%. This upgrade marks the evolution of AI agents from simple "single-task processors" to "digital assistants" with continuous cognitive capabilities. In the current industry context where the competition among large models is shifting from parameter scale to application deployment, the addition of long-term memory capabilities will significantly enhance the interaction depth and service continuity of agents in complex business scenarios.