Techniques of pure flooding queries through a peer-to-peer network[1] reach their limits for data localization which are not replicated and stored through a high number of nodes. To improve implicitly the query routing towards relevant nodes in order to smartly recover all the relevant data, we propose to organize a peer-to-peer network in which the logical neighborhood of each node is built according to the semantic of its content. Indeed, the data we target, are semantically rich, easily classifiable with a field of interest, or a theme (e.g. “rock”, “jazz”, ...in music, or “hydrology”, “oceanography”,...in sciences of environment). The motivation of our approach is based on the fact users/providers query the network on the theme of the data they store. Thus, if a node contains many data about the theme “Hydrology”, the query propagation is more efficient, if the neighbors of this node contain data about “Hydrology”. In the prototype VENISE (serVicE for Node Insertion in Semantic clustErs), we propose a protocol of node insertion based on cluster entries to cluster a peer-to-peer network.