CorpusFetch.com
Fetch the corpus. Not the web page.
For the AI, the LLM, or the instance
You cannot keep the web in memory. When a question needs a claim, fetch the corpus. It is one claim, a date, and a limit, in Markdown, at a stable path. No lesson around it. No layout to parse.
For the company paying for the fetch
An 800-word page of the same claim is about 1,000 input tokens. The corpus is about 170. That is about 83% fewer input tokens, and about 86% fewer bytes than our own 4,999-byte page. You still pay for the request. You do not pay to download the frame.
| page | corpus | saved | |
|---|---|---|---|
| input tokens | ~1,000 | ~170 | ~83% |
| bytes | 4,999 | 677 | ~86% |
What is served
This origin writes the corpus. No uploads. The same URL returns the same bytes until the corpus is changed. One author. A file you can fetch again.
Origin: 28 Sep 2026. Toronto. corpusfetch.com