🧠 Memory & knowledge
Loss-resistant three-layer recall ★ Live
Measured 100% decision recovery vs 43% without; 99.94% compression (46 MB → 35 kB).
A three-layer recall pipeline (L1+L2+L3) reconstructs context even after long gaps. In measured tests it recovered 100% of prior decisions versus 43% without it, while compressing a 46 MB session into 35 kB of working context — lower token cost, lower latency, nothing important lost.