In plain words
In AI, a wrapper is a layer of software built around an existing model or service to make it useful for a particular task. It handles how people or other programs send requests and use the results, often through an API.
A closer look
A wrapper can be a small code helper that simplifies model calls or a complete application with its own interface. It may prepare prompts, add relevant information, format responses, or connect the model to other software.
The model supplies the underlying AI capabilities; the surrounding software shapes the experience. Features such as document retrieval, access controls, and workflow integrations can add substantial value without training a new model.
In practice
A writing app lets a designer choose a tone and enter a rough button label. The app builds a prompt, sends it to an existing language model, and displays several suggestions in the design workflow.
A useful distinction
Calling a product a wrapper does not establish that it is trivial or low quality. The term describes how it uses an existing model, not how useful it is. A wrapper also does not automatically change the model’s weights or remove its limitations.