I don't understand a single line of code. I used AI to build INDXR MCP - my foundation for all projects. It indexes. I just chat with code. Shipped 19.4x on 1660S in 1 day without reading code.
Collecting feedback to improve INDXR MCP
#vLLM on my GTX 1660 served 12 tok/s single stream. People would've called it unusable.
After the same research loop that did the SGLang 21x: 67 tok/s single, 717 tok/s aggregate across 8 users.
Old GPU. New loops.
Indxr is an MCP server that turns AI coding agents into researchers. It indexes your repo, searches code and symbols, analyzes call graphs and hotspots, runs a full experiment loop — hypothesis, failure guards, design, measurement — plus a wiki that compounds lessons.