A method, computer program product, and a system where a processor(s), obtains, via a conversational agent executing on a client, a conversation thread and generates topic distributions. The processors(s) identifies topics and pre-fetches web content, based on utilizing the topics as search keywords. The processor(s) parses the pre-fetched web content to identify, in the pre-fetched web content, a first content type and a second content type. The processor(s) filters the parsed pre-fetched web content, to separate the first content type from the second content type and extracts a signature of a REST call associated with the pre-fetched web content from the first type of content and tokenizes the second type of content to supply context for invoking the REST call. The processor(s) generates a package that includes signature and the context and injects the package into the client so it is invocable by the conversational agent, based on the context.