MARTI enables training and inference of LLM-based multi-agent systems through reinforcement learning. The framework combines centralized multi-agent interactions with distributed policy training, supporting both built-in graph-based workflows and third-party multi-agent frameworks like AutoGen and CAMEL. MARTI-v2 extends the framework with tree search-augmented RL for complex reasoning tasks, incorporating advanced techniques like GSPO loss for sequence-level optimization and support for ultra-long sequences up to 32K tokens.