---
title: "Developer"
manual: "LLMs.txt & Markdown for AI"
version: "0.9"
permalink: "https://docs.typo3.org/permalink/woit/si-llms-txt:developer@0.9"
source: "Developer/Index.rst"
rendered: "2026-10-16T12:46:25+00:00"
---

# Developer {#developer}

## Request flow {#developer-request-flow}

The extension adds three PSR-15 middlewares to the frontend stack. No route
enhancer is involved: the middlewares answer the text files themselves and
remove `.md` before TYPO3 routes the request.

```text
request
  ├─ woit/si-llms-txt/markdown-suffix   before typo3/cms-frontend/site
  ├─   typo3/cms-frontend/site
  ├─ woit/si-llms-txt/llms-txt          before typo3/cms-frontend/base-redirect-resolver
  ├─   … page-resolver, prepare-tsfe-rendering, shortcut redirect
  ├─ woit/si-llms-txt/markdown-render   after prepare-tsfe-rendering
  └─   rendering
```

| Middleware | Task |
| --- | --- |
| `MarkdownSuffixMiddleware` | `/de/page.md` → routes `/de/page` and sets the request attribute `sillmstxt.markdown`. Responses without Markdown become a plain 404; redirects keep the `.md` suffix. |
| `LlmsTxtMiddleware` | Answers `llms.txt` / `llms-full.txt` for the site root and every language base. |
| `MarkdownRenderMiddleware` | Uses the page field "LLM Markdown" or lets the page render and converts the HTML. For the EXT:news detail page it takes the news record (see [EXT:news (tx_news)](https://docs.typo3.org/permalink/woit/si-llms-txt:news@0.9)). |

For normal HTML requests each middleware only compares a string and passes
the request on.

The event listener `AddMarkdownAlternateLink`
(`AfterCacheableContentIsGeneratedEvent`) adds the `<link rel="alternate">`
and `<link rel="describedby">` tags when a page is rendered; the result is
part of the page cache. The HTTP `Link` header with the same relations is
added by `MarkdownRenderMiddleware` on every request (see
[Discovery links](https://docs.typo3.org/permalink/woit/si-llms-txt:endpoints-discovery@0.9)).

## Classes {#developer-classes}

| Class | Purpose |
| --- | --- |
| `Service\LlmsTxtBuilder` | Builds `llms.txt`, `llms-full.txt` and the frontmatter of page and news documents; caching |
| `Service\PageCollector` | Walks the page tree, language overlay (strict) |
| `Service\NewsCollector` | News records of EXT:news (optional) |
| `Service\MarkdownConverter` | HTML → Markdown with league/html-to-markdown; removes navigation, breadcrumbs, asides, forms, buttons, scripts, decorative images and links without text; accordion titles and `<summary>` become h3 headings; keeps button titles in headings; resolves lazy-loaded images; moves card links onto their heading; makes links absolute |
| `Service\DiscoveryLinks` | `alternate` and `describedby` links of a page, as HTML tags or HTTP `Link` header |
| `Service\MarkdownUrl` | Page URL ↔ `.md` URL |
| `Service\SiteConfig` | Site set active? Settings, included page types |

## Caching {#developer-caching}

-   Cache `sillmstxt` (database backend, group `pages`)
-   Entries for `llms.txt` and `llms-full.txt` per site language
-   Tags: `pageId_<uid>` of every listed page and the root page;
    with EXT:news also `tx_news`, `tx_news_pid_<pid>`, `tx_news_uid_<uid>`
-   The intro of the language is part of the cache identifier
-   `llms-full.txt` with pages that could not be fetched is kept for
    5 minutes only
-   `.md` documents are not cached separately: the HTML comes from the page
    cache, only the conversion runs per request

## Database {#developer-database}

**ext_tables.sql**

```sql
CREATE TABLE pages (
    tx_sillmstxt_exclude smallint(5) unsigned DEFAULT '0' NOT NULL,
    tx_sillmstxt_priority int(11) DEFAULT '0' NOT NULL,
    tx_sillmstxt_description text,
    tx_sillmstxt_markdown mediumtext
);
```

`tx_sillmstxt_exclude` and `tx_sillmstxt_priority` use
`l10n_mode = exclude` (same value in all languages).