TrajectoryRL

About TrajectoryRL

TrajectoryRL is an agentic reinforcement learning (RL) platform operating as a Bittensor subnet (sn11), where miners compete to develop and optimize 'fusion policies'—programs that dynamically select and combine open-weight AI models to answer agent requests efficiently, safely, and reliably. The platform features continuous, stake-weighted competitions with validators evaluating submissions in controlled, verifiable environments. Winning policies are deployed for real-world use, and the protocol is evolving toward a fully open, verifiable environment registry for agent evaluation and reward. TrajectoryRL emphasizes cost-effectiveness, safety, and reliability in agent trajectories, providing a transparent, reproducible framework for agent optimization.