TencentCloud/TencentDB-Agent-Memory — reverse-engineered prompt

Build me a local first memory layer for AI agents in TypeScript for Node. I want it to help an agent remember things across both a single task and across many sessions, without sending anything to outside APIs.

The short term part should take big tool outputs and logs, save the full details locally, and keep only a compact symbolic view in context, ideally as a Mermaid style graph with traceable IDs so the agent can drill back into the raw text when needed. The long term part should turn conversations into structured memory, like raw chats, small facts, scenarios, and a higher level persona or profile, so the agent can reuse my preferences, workflows, and past context instead of making me repeat myself.

Please make it human readable and inspectable, with top level memory stored in markdown and lower level evidence retrievable from local storage. If it makes sense, include plugin support for OpenClaw and Hermes. Add a solid README, quick start, and a simple demo. You can look up current docs online if you need to.