EvoLib, introduced in the paper "Test-Time Learning with an Evolving Library," is a framework that enables large language models to learn from their own experience during inference without ground-truth labels or external feedback. It transforms past attempts into reusable skills and reflective insights, which are continually refined, consolidated, and reweighted, turning instance-specific observations into increasingly general knowledge over time. This approach helps AI models learn from past successes and failures and evolve knowledge with the highest potential for improving future performance. Unlike traditional AI memory systems that store raw experiences as static information, EvoLib extracts reusable knowledge and refines it as new experiences arrive, making skills more general and insights more accurate. Because EvoLib does not require model updates, it can be applied to any black-box language models and AI systems deployed through APIs. The framework addresses the limitation of memory alone, which is not learning, by identifying relevant knowledge for new tasks and refining it to improve performance consistently over time.