What is a context window?
A context window is how much text the model can see at once: your prompt, the files you attached, the earlier messages, and the answer it is writing.
Also called: context length
In plain English
It is the desk. If you pile every email from the year on the desk, the important page gets buried. When the window fills up, older instructions fall off. That is why a fresh chat with the right file beats a month-long thread. A long window is not a filing cabinet. Stuffing the whole year of emails into one chat makes the middle easy to skip, and you pay for every token that went in. Start the chat with the current job and attach only the thread that matters. When the chat gets long enough that it forgets the deposit, start a fresh one and paste the decision, not the whole argument.
How an owner would use it
Start a new chat per job. Attach the one SOP or the one spreadsheet tab that matters. Restate the rule you care about at the top, because the model will not reliably remember last Tuesday.
Common mistakes
A bigger window is not a memory and not a filing cabinet. Owners dump a year of email into one chat because the product allows it, then wonder why the rule from the first message disappeared. Long pastes also cost more usage and hide the one fact that mattered. Start a new chat for a new job. If you need last month, retrieve that page. Do not keep one immortal thread per client.
A worked example
A bookkeeper
One client, one chat
This is an illustration, not a customer result. The bookkeeper used to keep one giant chat per client. Numbers from March showed up in July's questions. They switched to one chat per close, pasted only that month's export, and put the 'do not guess a tax figure' line at the top.
Related terms
Prompts and guides
Questions owners ask
What is a context window?
It is the amount of text a model can consider at one time, including your instructions, attachments, and the conversation so far.
How do I use a context window without losing the plot?
Keep chats short and on one job. Attach only the file that matters. Repeat the hard rule in the prompt instead of hoping the model remembers it.
Will a long context window remember last quarter?
Only the text still inside the current conversation or the file you attached. When the window fills, older turns drop off. Saved memory, if the product has it, is a separate feature and goes stale on its own.
Product guides are on AI apps for business.