The number of information sources has dramatically enlarged lately, especially because of the growth of Internet, which has led to information overload of users. To use available information sources more efficiently, it is possible to enable user interface with intelligent agent paradigm - software components that are capable of performing tasks on behalf of other entities autonomously, with little or no influence. Intelligent agents are being executed within an information infrastructure and depend on a number of primitives that the infrastructure has to realize. Software environment for the execution of intelligent agents is in most cases built around a scripting language, providing means for serialization of agent state to and from external medium and its transfer to different node in the network. In that manner the independence of actual platform architecture is achieved. Often agents must interact with other agents and with various elements of information infrastructure. For that purposes one has to provide means of communication of high semantic level that will provide desired generality. Typical example is communication with an information broker that will enable access to knowledge related to particular field of interest. In this thesis possibilities for using RDF knowledge model have been considered. The model is based on knowledge representation using semantic networks and has been targeted for application in the field of Internet. Using appropriate syntax RDF statements can be inserted into XML documents. Apart from the knowledge model, means for manipulating knowledge are needed. A language for querying RDF databases and knowledge manipulation has been proposed, along with the formal query and inference model. The syntax for serializing RDF queries into XML documents has been provided. Proposed models and languages have been tested by implementing an information broker based on RDF and RDF query language. System architecture and means of realization have been given in the thesis.