This paper focuses on two machine learning abstractions springing from ecological models: (i) evolutionary adaptation by local selection, and (ii) selective query expansion by internalization of environmental signals. We first outline a number of experiments pointing to the feasibility and performance of these methods on a general class of graph environments. We then describe how these methods have been applied to the intelligent retrieval of information distributed across networked environments. In particular, the paper discusses a novel distributed evolutionary algorithm and representation used to construct populations of adaptive Web agents. These {\em InfoSpiders} search on-line for information relevant to the user, by traversing hyperlinks in an autonomous and intelligent fashion. They can adapt to the spatial and temporal regularities of their local context. Our results suggest that InfoSpiders could complement current search engine technology by starting up where search engines stop. Engines provide global starting points, based on statistical features of the search space (words); InfoSpiders can use topological features (links) to guide their subsequent search on-line. We show how this approach can help overcoming some of the limitations of the current state of the art, dealing with the problems of scalability, personalization, and localization.