Find official introductory documentation explaining what a language-model context window is.
Return two source links, each with the specific claim it supports.
Separate provider limits from general definitions. Do not invent a universal token limit.