How I built a zero-token memory layer for LLMs (and why it outperforms vector store approaches)

If you've built an AI chatbot or agent, you've hit the same problem: the LLM forgets everything...