Early Access · Shaping the first primitive

The setup layer for frontier RL.

Design RL environments, tasksets, and eval loops without rebuilding the setup stack from scratch. Models are getting good enough that the painful part of RL is no longer only training.

STEP 01intentresearch idea
STEP 02env + taskswhat to train on
STEP 03rewards + evalswhat counts as good
STEP 04trainable loopship it