microsoftthinkingboxthinkingbox is a framework for defining tool as MCP servers, running LLM agents against them, and evaluating agent behavior — for offline training-data generation, reinforcement-learning training loops, and model evaluation.Take a look →
nvlabsalpagymAlpaGym is a reinforcement-learning framework for end-to-end autonomous-driving policies.Take a look →
thudmscale-cuaOpen-source framework for computer use agents: VeriGen verifiable task synthesis, online RL training (AgentRL), and OSWorld/ScienceBoard evaluation.Take a look →
xyz-ai-labaxrlAxisRL is an agentic RL post-training framework built on SGLang rollout, Megatron training, and real-world agent workflows.Take a look →
