This paper presents a Multi-Agent based Data Integration (MADI) framework for integrating distributed data source across Internet. This framework takes control on high-availability and high performance without compromising the data integrity and security. Special efforts on agent identity management and task scheduling and collaboration strategy were employed to guarantee above assertion. And a built-in Agent Frontier Data Cleaning mechanism can strengthen data integrity and boost up performance. Moreover, this approach is optimized especially for working under coarse network and runtime environment. The results from lab experiment and field test will fortify the claims.