Reglos: Self-Evolving Personal Memory Agent via Recursive Improvement

Reglos is a training-free agent that uses a recursive self-improvement loop to update its memory harness. A coding agent proposes patches based on failure traces, while a second model family acts as a critic to prevent benchmark gaming and overfitting.