wtf is tencentdb-agent-memory?

TencentDB Agent Memory is a memory system for AI agents built by TencentCloud. The idea is simple: instead of making an AI assistant repeat the same background information every time you talk to it, this project lets the assistant store and reuse what it has already learned, such as your project setup, your preferred way of working, and past conversations. The project describes two main ideas behind how it stores memory. The first is layering. Rather than dumping everything into one flat pile of data, information is organized in levels: raw conversation at the bottom, then extracted facts, then summarized scenes, and finally a top level persona that captures a person's habits and preferences. The system only needs to look at the top level most of the time, and can drill down to the more detailed levels when something needs checking. The same layering idea is applied to skills the agent picks up over time, turning repeated actions into reusable routines. The second idea is symbolic memory. Long tasks tend to generate huge amounts of text such as search results, code, and error logs. Instead of keeping all of that in the assistant's active context, the full details are saved to separate files, while a compact diagram called a Mermaid canvas keeps track of how the task is progressing. If the assistant needs to check a detail later, it can look up a reference id and pull the original text back up. According to the README, when used alongside a tool called OpenClaw, this approach reduced token usage by over sixty percent on one benchmark and noticeably improved task success rates and memory accuracy on others. The project is written in TypeScript, distributed as an npm package, and requires Node.js version 22.16 or higher. It is released under the MIT license, which allows free use including commercial use.