Tencent Open-Sources TencentDB Agent Memory for AI Agent Memory Management
Tencent Cloud has open-sourced TencentDB Agent Memory, a hierarchical memory engine designed to help AI agents structure conversations, documents, and code into reusable memory assets. The TypeScript-based repository has gained significant traction, accumulating over 3,600 GitHub stars in a single week. Memory management is a critical bottleneck in multi-agent systems, and this tool addresses it by reducing token consumption by up to 61.38% while improving task completion rates by 51.52%. It allows teams to share and govern memory assets across different agents and frameworks, moving from stateless tools to collaborative assistants. The system organizes memory into four distinct assets: Chat Memory, Skill, LLM-Wiki, and Code-Graph. Released under the MIT license, it offers out-of-the-box support for OpenClaw and Hermes Gateway.
## BACKGROUND
AI agents often operate statelessly, meaning they treat each prompt or task independently without retaining context from past interactions. Effective memory management allows agents to store long-term embeddings and short-term reasoning context, enabling them to recall information across sessions and handle complex, long-running workflows without information overload.