Context window
A context window is the maximum amount of text, measured in tokens, that a language model can process in a single request — the working memory that must hold the system prompt, the conversation, fetched documents, and tool results all at once. Everything an agent knows about an API at decision time has to fit inside it, alongside everything else the task requires.
The window is a budget, and API documentation competes for it. An agent mid-task is holding its instructions, the user's request, the code it is writing, and possibly several other services' docs; a single API's documentation gets a slice, never the whole. Docs that arrive as concise markdown spend that slice on content, while docs that arrive as rendered HTML spend much of it on markup and navigation.
Producers feel this most at the extremes. A full-text bundle like llms-full.txt is only loadable if its size is sane relative to real windows, and long reference pages with the answer buried under preamble force the agent to spend tokens reaching it. Documentation written for a reader with unlimited patience meets a reader with a hard cap.
Discry's comprehension dimension includes a token-efficiency check — whether the docs are concise enough to fit an agent's context window — and the llms-full.txt discovery check reports the bundle's token count, so the cost of a full ingest is visible on the profile before anyone pays it.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free