Last released Sep 15, 2026
Self-improving agents through continual learning: in-process agentic RL that turns your own agent's work into training data. Observed tool outcomes, outbound-only, one runtime dependency.