The paper reviews the development of generic user modeling systems over the past twenty years. It describes their purposes, their services within user-adaptive systems, and the different design requirements for research prototypes and commercially deployed servers. It discusses the architectures that have been explored so far, namely shell systems that form part of the application, central server systems that communicate with several applications, and possible future user modeling agents that physically follow the user. Several implemented research prototypes and commercial systems are briefly described. User modeling systems were first developed in the 1970s, with early work by Allen, Cohen, Perrault, and Elaine Rich. These systems allowed applications to collect and adapt to user data. In the mid-1980s, systems began to separate user modeling components from other system components. In 1986, Tim Finin introduced GUMS, a general user modeling system that allowed for the definition of user stereotypes and rules for reasoning about them. This system laid the foundation for future user modeling systems. In the 1990s, several research groups developed user modeling shell systems, including UMT, BGP-MS, DOPPELGÄNGER, TAGUS, and um. These systems provided various services such as representing user assumptions, classifying users into stereotypes, recording user behavior, and maintaining consistency in user models. However, these systems were often limited in their expressiveness and inferential capabilities. The late 1990s saw the beginning of a commercial boom in user modeling systems, driven by the increasing value of web personalization in e-commerce. Commercial systems such as Group Lens, LikeMinds, Personalization Server, Frontmind, and Learn Sesame were developed to support personalized web experiences. These systems often used client-server architectures, allowing for the maintenance of user information in a central repository and enabling multiple applications to access it. Commercial user modeling systems differ from academic ones in that they must support services such as user comparison, external data import, privacy support, and quick adaptation. They are also more focused on practical applications and less on theoretical aspects. However, they often lack the generality, expressiveness, and inferential capabilities of academic systems. The future of user modeling systems is likely to involve mobile user models and user models for smart appliances. These systems may be implemented as user model agents that follow the user or are integrated into devices such as cars and home appliances. Additionally, user modeling systems may be used for multiple purposes, including organizational directory services and expert finding systems. The need for standardization in user modeling systems is also expected to grow, driven by privacy requirements and the need to integrate user information across different systems.The paper reviews the development of generic user modeling systems over the past twenty years. It describes their purposes, their services within user-adaptive systems, and the different design requirements for research prototypes and commercially deployed servers. It discusses the architectures that have been explored so far, namely shell systems that form part of the application, central server systems that communicate with several applications, and possible future user modeling agents that physically follow the user. Several implemented research prototypes and commercial systems are briefly described. User modeling systems were first developed in the 1970s, with early work by Allen, Cohen, Perrault, and Elaine Rich. These systems allowed applications to collect and adapt to user data. In the mid-1980s, systems began to separate user modeling components from other system components. In 1986, Tim Finin introduced GUMS, a general user modeling system that allowed for the definition of user stereotypes and rules for reasoning about them. This system laid the foundation for future user modeling systems. In the 1990s, several research groups developed user modeling shell systems, including UMT, BGP-MS, DOPPELGÄNGER, TAGUS, and um. These systems provided various services such as representing user assumptions, classifying users into stereotypes, recording user behavior, and maintaining consistency in user models. However, these systems were often limited in their expressiveness and inferential capabilities. The late 1990s saw the beginning of a commercial boom in user modeling systems, driven by the increasing value of web personalization in e-commerce. Commercial systems such as Group Lens, LikeMinds, Personalization Server, Frontmind, and Learn Sesame were developed to support personalized web experiences. These systems often used client-server architectures, allowing for the maintenance of user information in a central repository and enabling multiple applications to access it. Commercial user modeling systems differ from academic ones in that they must support services such as user comparison, external data import, privacy support, and quick adaptation. They are also more focused on practical applications and less on theoretical aspects. However, they often lack the generality, expressiveness, and inferential capabilities of academic systems. The future of user modeling systems is likely to involve mobile user models and user models for smart appliances. These systems may be implemented as user model agents that follow the user or are integrated into devices such as cars and home appliances. Additionally, user modeling systems may be used for multiple purposes, including organizational directory services and expert finding systems. The need for standardization in user modeling systems is also expected to grow, driven by privacy requirements and the need to integrate user information across different systems.