3 ms·
ChatGPT agrees with you too much. I built an AI that pushes back
- schmommy 8mo agoYou know that feeling when you ask ChatGPT about your architecture decision and it's like "That's a great approach!" even though you're about to shoot yourself in the foot? I got tired of that. So I built MIMIR — an AI that runs 2-4 models in parallel on every message. A Lead Analyst breaks down your problem. A Strategic Advisor thinks long-term. A Creative finds the angle you missed. Then a synthesis layer merges their thinking into one response. Here's what's different: When you float a bad idea, MIMIR tells you it's bad. When there's a tradeoff you're ignoring, it surfaces it. When you're optimizing for the wrong thing, it pushes back with a better path. Not because I programmed it to be contrarian — but because multiple models thinking independently means blind spots get caught before they become your answer. The stack: Multi-model collective intelligence (2-4 models per response) Persistent memory across sessions (not context window tricks) Self-improving: learns from corrections in real-time Can query your connected databases and services via MCP tools