RLM+Ralph is a low-cost autonomous coding loop: each iteration plans → writes code → verifies → repeats. RLM keeps context small by using external memory + file indexing, so cheap models can iterate indefinitely without context overflow, correcting mistakes until done.