I spent the last few months building an agent that doesn't just write code, but optimizes it through a closed-loop sandbox. It uses Gemini for high-level strategy and Groq/Llama-3 for surgical HPC implementation. It iterates, benchmarks itself, and breaks its own records until it reaches the theoretical limits of the hardware.

Today, I’m out of resources. The "compute wall" is real for an independent dev. Instead of letting this rot on my drive, I’m releasing it.

A note on the license (ECO-AGENT APEX-PROTOCOL): I’ve spent my last credits on this. If you use this for science, education, or to make the world better, it’s yours for 8% royalties (only if you make money). If you use it for high-frequency trading, mass surveillance, or to fuel the chaos of the world, I’m taking a 30% "Chaos Tax". I trust you to self-assess.

The tomorrow we build is made of the acts of the past. Be careful with what you’ll find in the vault.

Source: https://github.com/KilianDiama/AutonomousRDAgent