We propose a dynamic cooperation model for different agents' message exchanging. Besides, we propose a new algorithm, which makes it easy to transform a traditional Boolean logical query statement into a matrix. This method allows a document might be retrieved in a clear Boolean view even it matches the query partially. On the base of them, we implement an agent-based selective information retrieval. Parallel crawlers make the prototype scalability and network-load reduction. The focused crawler makes it easy to find as many relevant pages and as few irrelevant pages as possible. Through experiments, we show the results.