Perpetual Machines

High-fidelity training data for AutoResearch.

Non-hackable RL environments. Private task data. Hidden solutions. Thousands of procedurally generated, verifiable variants grounded in real work.

We're building for:

Frontier model labs — high-fidelity environments for RL and evaluation.
ML/AI teams — deploying 24/7 AutoResearch systems that autonomously improve against any eval.