MemTensor/MemOS — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a self hosted memory system for AI agents, like an operating system for memory. I want agents to remember useful facts, past conversations, tool use, documents, images, and user preferences across sessions, then bring the right memories back when chatting or doing tasks.

It should have simple APIs to add, search, update, correct, and delete memories. I also want separate memory spaces for different users, projects, or agents, with a way to share selected knowledge between them. Make it possible to give natural language feedback, like “that memory is wrong” or “replace this with the new info,” and have the system improve over time.

Please include a local first plugin style setup for agent tools, plus a basic dashboard or viewer so I can inspect and manage memories instead of treating it like a black box. Aim for lower token usage by reusing stored context instead of sending everything every time. Look up current docs online if you need to.

Want more depth? Deep Reverse