What a context window is
The context window is the maximum amount of text — measured in tokens — that a model can take into account for a single response. It includes your prompt, the conversation so far, retrieved code, and tool outputs.
Everything the model 'knows' in the moment lives inside this window. Anything outside it simply isn't considered.
What happens when it fills up
As a session grows past the window, earlier content gets summarized or pushed out. The model doesn't warn you — it just stops accounting for the constraint you set far enough back.
This is why long coding sessions drift: the agent reintroduces a rejected pattern because the original reasoning fell out of the window.
Bigger windows: help and limits
Larger context windows delay the drop-off and let you include more at once. But they cost more, add latency, and dilute attention — stuffing in everything can make answers worse, not better.
A bigger window is more room, not more focus. You still have to choose what deserves the space.
Window is not memory
The window resets every session. A huge window doesn't make the model remember your project tomorrow — that requires persistent memory you re-supply, like a maintained project file.
Conflating 'big context' with 'memory' is the most common misconception, and the source of a lot of frustration.
Budget it deliberately
Treat the window as a budget. Spend it on what changes the answer — constraints, the relevant code, recent decisions — and trim transcripts and boilerplate.
Keep durable facts in persistent memory and pull task-specific detail via retrieval, so the window stays focused.