The open source post-building layer for agents. Our environment data and evals power agent post-training (RL, SFT) and monitoring.
Terminal score: 0–1000 raw, weighted across 4 dimensions. Public score: 0–10 normalized (shown in the 30-day stars chart above).
