"让 Token 消耗降低 61%"：腾讯开源 Agent Memory

AIMPACT News, May 14 (UTC+8), Tencent Cloud open-sourced TencentDB Agent Memory, aiming to address the issues of context window overflow and high token costs in long agent tasks. The solution adopts two core technologies: “Context Offloading” and “Mermaid Task Canvas,” which offload complete information to external storage while preserving key states and execution paths with a structured task diagram. Experiments show that this approach can reduce token consumption by up to 61% in multi-task continuous conversations and improve task success rates. The project has been integrated with mainstream frameworks such as OpenClaw, supporting one-click integration and local SQLite storage. (Source: AiHot)