AgFlow is a middleware platform for quality-of-service (QoS)-aware Web service composition. It addresses the challenge of selecting Web services for composition to maximize user satisfaction based on QoS attributes while meeting user and service constraints. The platform uses two service selection approaches: local optimization, which selects services for individual tasks without considering inter-task QoS constraints, and global planning, which uses integer programming to optimize service selection across the entire composite service. AgFlow's architecture includes a service broker for registering services, an execution planner for generating execution plans, and an adaptive execution engine that adjusts plans based on runtime changes. The platform features a multidimensional QoS model capturing non-functional properties like availability and reputation, and an adaptive execution engine that replans when component services change or become unavailable. AgFlow is designed for service-oriented architectures where QoS is critical due to the distributed and dynamic nature of services. The system uses statecharts to model service dependencies and provides aggregation functions for computing QoS metrics for composite services. The paper presents a detailed description of AgFlow's architecture, service ontologies, execution paths, and QoS model, along with two service selection approaches: local optimization using multiple criteria decision making (MCDM) and global planning using integer programming. The system is evaluated through implementation and experimental results, demonstrating its effectiveness in selecting Web services for composition based on QoS criteria.AgFlow is a middleware platform for quality-of-service (QoS)-aware Web service composition. It addresses the challenge of selecting Web services for composition to maximize user satisfaction based on QoS attributes while meeting user and service constraints. The platform uses two service selection approaches: local optimization, which selects services for individual tasks without considering inter-task QoS constraints, and global planning, which uses integer programming to optimize service selection across the entire composite service. AgFlow's architecture includes a service broker for registering services, an execution planner for generating execution plans, and an adaptive execution engine that adjusts plans based on runtime changes. The platform features a multidimensional QoS model capturing non-functional properties like availability and reputation, and an adaptive execution engine that replans when component services change or become unavailable. AgFlow is designed for service-oriented architectures where QoS is critical due to the distributed and dynamic nature of services. The system uses statecharts to model service dependencies and provides aggregation functions for computing QoS metrics for composite services. The paper presents a detailed description of AgFlow's architecture, service ontologies, execution paths, and QoS model, along with two service selection approaches: local optimization using multiple criteria decision making (MCDM) and global planning using integer programming. The system is evaluated through implementation and experimental results, demonstrating its effectiveness in selecting Web services for composition based on QoS criteria.