EnvACE is a new agentic reinforcement learning method that internalizes environment dynamics by replacing external environment interaction during training with world rehearsal. This allows for strong and transferable performance across various benchmarks, including FinMCP-Bench.