This paper presents an information agent and latent semantic-based indexing architecture to retrieve documents on the Internet. The system optimizes the search for documents in the Internet by automatically retrieving relevant links. The information used for the search can be obtained, for instance, from Internet browser caches and from grades of relevance manually informed. To leverage the scope of retrieved documents, the system makes use of existing indexing mechanisms. Returned documents are then filtered using Latent Semantic Indexing (LSI). In a co-operative environment, the proposed architecture provides for sharing of documents and grades among the group. The architecture has been used in a cooperative learning environment, where students share their browser caches and retrieved documents. The paper focuses on the architecture of the information agent; the context reconnaissance, search, and filter modules are described. Scenarios of usage and system performance are also addressed. Three figures present the architecture of the agent, the keywords generation process, and a mathematical representation of the decomposition into singular value used to implement LSI. (Author/AEF)