Context Window

In one sentenceHow much an AI can "keep in mind" at once: your instructions, pasted documents and the conversation so far.

What it means

Every model has a limit on how many tokens it can consider in one go. Everything counts toward it: your prompt, any files you attach, previous messages in the chat, and the reply it writes.

Modern models have large context windows (hundreds of thousands of tokens, some over a million), enough for long reports or entire books. But even within the limit, very long chats can get fuzzy, and early details can get less attention.

How to use it

• For a new topic, start a new chat instead of continuing a 200-message thread.
• Put the most important instructions at the start, and repeat key constraints at the end of long prompts.
• When the AI "forgets" something you said earlier, paste it again or summarize the chat and start fresh.

Related terms

Token Prompt Memory Context Engineering