volcengine/OpenViking ? reverse-engineered prompt

I want to build a context database specifically for AI agents. The main idea is to manage an agent's brain—its memory, skills, and resources—using a file system paradigm. Instead of just having a messy vector database, I want to organize everything into a clear, hierarchical structure like folders and files on a computer.

This system should be really efficient with tokens, using a tiered loading approach so the agent only pulls the context it needs at that moment. The agent should be able to search this "file system" by combining directory navigation with semantic search to get really precise results. It's also important to be able to visualize the agent's retrieval path to make debugging easier.

Finally, I want it to have automatic session management that can compress conversations and extract long-term memories, so the agent actually gets smarter as it's used. Make sure it can connect to various LLM providers like OpenAI, Anthropic, and others.