<p>A federation is the interaction between different systems to facilitate the exchange of information between various points. For instance, a parking search in which nodes located in different parking lots interact. In the Web of Things domain, discovery services interact with each other to exchange information about nearby devices, allowing users to access devices deployed elsewhere. These device searches use syntactic, semantic, and natural language search techniques through Artificial Intelligence (AI) models. The natural language search technique in the federation currently works with a single trained model that contains all the information from the nodes in the network. In this paper, we propose an extension of that work by decentralizing artificial intelligence models in natural language search. Instead of having a single artificial intelligence model on the cloud for all the nodes, one model is deployed for each node in the federation, trained only using the data managed by that node. This solution aims to improve the accuracy of AI models by moving the centralized AI from the cloud to a layer closer to the devices, the edge layer. Therefore, the infrastructure moves from a centralized continuum computing solution to a distributed AI for edge solution. After experimentation, the proposal was validated by comparing the performance of the federation with a centralized AI model against the federation with a per-node AI model. The results show that the edge-distributed solution solves queries faster than the centralized approach and has more flexibility in defining the search space.</p>