# AI access

Portable Markdown exports for humans and language models.

## Static exports [#static-exports]

The build creates `llms.txt`, `llms-full.txt`, `ai-manifest.json`, and physical
`docs/**/*.md` Markdown twins under `apps/portfolio/out/`. Each manifest entry maps
an HTML page to a Markdown file and contains the text Fumadocs processed from
its MDX source. The exporter does not use a second Markdown converter. No runtime
API or rewrite service is required.

## Crawl and syndication artifacts [#crawl-and-syndication-artifacts]

Next.js generates `robots.txt` and `sitemap.xml`; `tools export-ai` writes
`feed.xml` (RSS 2.0) and `feed.json` (JSON Feed 1.1). The sitemap lists the
home, search, and packages routes and every documentation page. Robots allows
all crawlers, including the named AI crawlers, and points to the sitemap.
Undated pages omit dates in the sitemap and feeds instead of inventing them.

## Generate and verify [#generate-and-verify]

```sh
pixi run bun run --cwd apps/portfolio build
pixi run bun run --cwd apps/portfolio test:seo
pixi run tools docs-link-check --offline --site-dir apps/portfolio/out
```

The homepage footer and documentation header link to `llms.txt`. Every docs
page also links to its Markdown twin and advertises it in its HTML metadata.
The site is rooted at `https://archont561.github.io/`; there is no base-path
setting. Pagefind searches the exported HTML in production. If custom MDX
components are introduced, provide a portable Fumadocs Markdown representation
and a regression test.