Test-time prompt learning now works in real-world multi-dataset scenarios through intelligent task routing and co-evolution—enabling agents to adapt their behavior to diverse, heterogeneous data streams without retraining.
EEVEE is a framework that lets AI agents improve themselves during real-world use by learning better prompts on-the-fly. Unlike existing methods that work on single datasets, EEVEE handles messy real-world data from multiple sources by routing different types of tasks to specialized prompts, then co-evolving both the router and prompts together.