MemVault: Building a 3-Tier LLM Memory System That Cuts Token Costs by 56%

I built MemVault, a complete, working 3-tier memory architecture for LLM applications that reduces token costs by 56% while giving AI persistent knowledge about users across sessions.
Live demo will show: the real-time cost dashboard tracking every token, the interactive D3.js graph visualizing exactly what the AI “knows” about a user, the smart model router switching between models automatically, and the full Redis, PostgreSQL, ChromaDB memory pipeline in action.