Open-source AGI R&D platform enabling agents for benchmark and open-ended tasks. Powered by GPT-4, Vicuna, LLaMA, Flan-T5. Supports multi-modality (text, image, video, audio) tool learning and task solving. Features linear and non-linear task decomposition, benchmark and open-ended task solving, easy evaluation protocols, and RLTF (Reinforcement Learning from Task Feedback) for continuous self-improvement.