The Context Window

You tell an assistant the time of an event. Ten messages later, it uses that time in an invitation. It seems to remember.

The application makes that continuity possible. It keeps the conversation and supplies relevant information when asking the model for another reply. The trained weights provide learned abilities; the request provides the working details.

Those details can include more than the visible chat. App instructions, documents, and saved notes may all be part of the input. This supplied material is the context. The context window is the limited space available for it.

A desk makes a useful comparison. You can put the current brief beside the assignment, but you cannot spread an unlimited archive across it. The app may shorten a long conversation, select passages, or ask you to reduce the input. A summary can lose an important requirement.

Even material that fits can be missed. A large working space does not guarantee careful use of every detail.

The practical habit is to keep the current facts together, preserve unknowns, and check that the answer used what mattered. The remembering may feel mysterious from the chat box. Following the request makes it understandable.