This paper introduces Antares, a bio-inspired algorithm that exploits ant-like agents to build a P2P information system in Grids. The work of agents is tailored to the controlled replication and relocation of metadata documents that describe Grid resources. These descriptors are indexed through binary strings that can either represent topics of interest, Specifically ill the case that resources are text documents, or be the result of the application of a locality preserving hash function, that maps, similar resources into similar keys. Agents travel the Grid through P2P interconnections and, by the application of ad hoc probability functions, they copy and move descriptors so as to locate descriptors represented by identical or similar keys into neighbor Grid hosts. The resulting information system is here referred to as self structured, because it exploits the self-organizing characteristics of ant-inspired agents, and also because the association of descriptors to hosts is not pro-determined but easily adapts to the varying conditions of the Grid. This self-structured organization combines the benefits of both unstructured and structured P2P information systems. Indeed, being basically unstructured, Antares is easy to maintain in a dynamic Grid, in which joins and departs of hosts can be frequent events. On the other hand, the aggregation and spatial ordering of descriptors call improve the rapidity and effectiveness of discovery operations, and also enables range queries, which are beneficial features typical of structured systems.