Aether AI has unveiled the RSIAgent framework, which leverages an innovative experience scaling strategy that eliminates the need for updating model parameters. Instead, it fosters recursive self-improvement by empowering agents to autonomously venture into and explore uncharted environments. RSIAgent orchestrates a multi-agent recursive closed-loop system, comprising a curriculum agent, an actor agent, and a verifier agent. It adopts a two-stage self-exploration strategy: initially conducting extensive exploration to construct a cognitive map of the environment, followed by a targeted approach to tackle key challenges. Throughout this process, stable causal relationships between actions, conditions, and outcomes are meticulously distilled and stored in Memory for future task utilization. Experimental findings reveal that RSIAgent substantially boosts the agent capabilities of open-source models like Kimi-K3 and GLM-5.3. In rigorous tests such as OSWorld 2.0 and Agents’ Last Exam, it surpasses the performance of closed-source models like GPT-6 Astra and demonstrates adaptability to diverse fields, including game development. This accomplishment marks a significant stride in the causal-driven agent methodology, propelling AI from a state of passive training to one of active causal discovery and learning. Looking ahead, the ceiling of agent capabilities will increasingly hinge on the causal knowledge garnered through active exploration.