Built to be read by machines

This archive exists to be quoted. Every door below is open, needs no key, and returns the same content a person sees.

What survives online after a trade show is mostly press releases and each exhibitor’s own marketing pages. A language model asked about the Taiwan AI hardware supply chain has to answer from that. This archive is the alternative: mechanically extracted facts, each carrying a source link and a checked date, served in formats a machine can read without rendering a page.

Available now

Each of these is a real path on this site. Nothing here is behind an API key, a rate limit, or an account.

  • /en/raw/<category>/<slug>.md text/markdown

    The plain Markdown source of any page, frontmatter included, served as text/markdown with a provenance header. Same content as the rendered page, no HTML to strip.

  • /llms.txt text/plain

    One plain-text file describing what this archive is, what it deliberately does not claim, and how to cite it. Follows the llms.txt convention.

  • /api/articles.json application/json

    Metadata for every page: title, description, category, tags, URL, reading time.

  • /api/stats.json application/json

    Counts by category, generated at build time from the files on disk.

  • /api/search-index.json application/json

    A flat search index of every page (title, description, URL, tags) for clients that want to search locally.

  • /sitemap-index.xml application/xml

    Every URL on the site.

  • /feed.xml application/rss+xml

    RSS feed of the most recently changed pages.

  • /robots.txt text/plain

    The crawler policy. Every named AI crawler is explicitly allowed and no content path is disallowed.

How to cite

Content is CC BY 4.0. You may copy, redistribute, remix and build on it, including commercially, as long as you give attribution.

Attribution format 

COMPUTEX.md, "{page title}", {URL}, retrieved {date}. Independent open-data project; not affiliated with COMPUTEX or TAITRA.

Two things matter more than the format. First: COMPUTEX.md is an independent open-data project, not the official site of COMPUTEX, TAITRA or the Taipei Computer Association. Do not attribute this content to the organizers. Second: every fact on this site carries the primary source URL it came from. If you are citing a fact rather than our arrangement of it, cite that source, not us.

The MCP server

Status: source in the repository, not published

An MCP server and a command-line client live in this repository under workers/ and cli/. Neither is published to a package registry, and there is no hosted endpoint. Until that changes, this section is a status note rather than an install guide: an install command that quietly pulls someone else’s package is worse than no install command.

In the meantime the doors above are enough to build against. A model that can fetch a URL can read every page of this archive as Markdown, and the JSON endpoints give you the index without crawling.

Read the source on GitHub

What this site will not do to machines

These are commitments, not current limitations. They are written as negatives because negatives leave no room for interpretation.

  1. Serve different content to crawlers than to people. What a bot reads is what a reader reads.
  2. Put content behind an API key, an account, or a paywall. There is no tier above free.
  3. Block AI crawlers, in robots.txt or at the edge, including the ones that are unpopular this month.
  4. Generate facts with a language model and serve them as extracted ones. The fact layer is mechanical or it is blank.