Context window

The context window is the total text budget available to a model for one request. The user question, system instructions, chat history, retrieved passages, and generated answer consume different parts of that budget.

When the window is full, the system cannot keep adding text. Retrieving relevant passages is more manageable than sending an entire document with every question.

RAG does not try to compress a whole archive into one context. It selects passages close to the question and moves those passages into the prompt, so a relevant clause can be processed instead of an entire eighty-page procedure.

Selection is not perfect. If retrieval returns noise, the context fills with the wrong material; citations and an explicit no-answer behaviour help manage that boundary.

Headings, numbered clauses, and short sections create meaningful chunk boundaries. Very long tables or unrelated topics placed in one file can make the relevant passage harder to select.

For users, the best approach is to ask within the scope of the appropriate assistant and name a relevant document when needed. A larger capacity does not by itself fix poor scope or stale files.

A useful way to evaluate context-window is to follow one real question from the source document to the final answer. Record which file was selected, what context reached the model, and what a reviewer would need to verify. This turns a definition into an operational check and makes the result comparable across teams.

The same check should include ownership and change management. Decide who updates the relevant documents, how an outdated result is reported, and which access boundary applies. KobiGPT can provide the assistant and the cited document context, but the organisation still owns the source material, permissions, and the decision made from the answer.