AgileRL

AgileRL builds tools for training AI agents with reinforcement learning. Its open-source framework offers evolutionary hyperparameter optimisation that the company says speeds up training by up to ten times compared with standard RL libraries. On top of this sits Arena, an enterprise platform that turns a customer's own data into training environments and datasets, runs distributed reinforcement fine-tuning of open models, and deploys the resulting agents on the customer's infrastructure. The company came out of Entrepreneur First, has offices in London and San Francisco, and has partnered with NVIDIA on post-training Nemotron models.