LLMify

Fixed

• Fixed an error at the end of craft update when updating from 1.x, after the update had already succeeded. (#35)

This release contains breaking changes. Please review breaking changes in the upgrade guide.

Added

• Add the cacheDuration plugin setting (default 86400 seconds), the longest time a page's markdown stays cached. 0 keeps it until it gets invalidated. Set it in config/llmify.php.
• Add an LLMify markdown option to Utilities > Clear Caches, also available as php craft clear-caches/llmify.
• Pages whose markdown could not be rendered are now listed on the dashboard, and the element sidebar shows the reason (e.g. a template error, no {% llmify %} blocks, empty markdown, or a failed front end request in headless mode).

Changed

• Markdown is now rendered on demand and kept in Craft's data cache, instead of being generated by queue jobs and stored in the database. The cache uses Craft's element cache tags, so a page's markdown is invalidated as soon as the page or anything it shows (related entries, assets, globals) changes, no matter where it is saved from. Previews are never cached.
• In headless mode, the front end is fetched on the first request of a page and the result is cached for cacheDuration.
• Saving the plugin, site or content settings now clears all markdown caches.
• php craft llmify/markdown/generate now warms the cache by caching every page that is not cached yet, e.g. nightly on sites under heavy load. php craft llmify/markdown/clear clears it.
• The utility shows per site how many pages are cached and since when. Its actions are now Warm Cache and Clear Cache, and the permissions are now called Warm Markdown Cache and Clear Markdown Cache.
• The dashboard shows how many pages are cached, the average tokens per page and the oldest cache entry. Its markdown score now counts the pages without failures.
• The element sidebar shows whether a page is cached, since when and its size in tokens. Refresh renders the page again, Clear removes it from the cache.

Removed

• Remove llms-full.txt, the Enable llms-full.txt site setting, the Include Front Matter in llms-full.txt plugin setting and the llms-full-txt API endpoint. The file is not part of the llms.txt spec, large sites easily outgrow any context window with it, and AI crawlers barely request it. Agents find every page's markdown through llms.txt and the .md URLs instead.
• Remove the llmify_pages database table. Stored markdown is dropped on update and rendered again on the next request.
• Remove RefreshService (Llmify::getInstance()->refresh), the RefreshMarkdownJob queue job and the X-Llmify-Refresh-Request header. Use Llmify::getInstance()->markdown->getPageMarkdown() to get a page's markdown.

Fixed

• Markdown no longer depends on the visitor who requests a page first: the template renders without their query string, cookies, User-Agent and Accept-Language, and cookies it sets do not end up in the markdown response.
• Markdown served through content negotiation now sends Cache-Control: private, no-store, so CDNs that ignore Vary (e.g. Cloudflare) do not serve it to browsers.
• Resolving seomatic: values no longer leaves SEOmatic in its preview state for the rest of the request.

Changed

• Refresh the bundled AI bot list.

Fixed

• Links in generated markdown no longer carry a siteToken parameter on live sites. (#32)

Added

• Add rel="describedby" discovery for the llms.txt spec v2: the discovery tag now also injects <link rel="describedby" href="/llms.txt">, and .md responses (direct and content-negotiated) send a Link header pointing to llms.txt (rel="describedby") and, for negotiated responses, to the .md URL (rel="alternate"; type="text/markdown").
• Show a warning on the dashboard while a Markdown URL Prefix is set, since the spec expects markdown at the original URL with .md appended.

Changed

• The homepage markdown is now served at /index.md instead of /__home__.md, as the spec requires for URLs without a file name. Generated links, the discovery tag and llms.txt use the new URL; __home__.md keeps working.
• Markdown URL Prefix now defaults to an empty string for new installs, so .md URLs match the spec. Existing installs that never saved the setting keep raw via a migration.

Added

• Add Basic Auth Username and Basic Auth Password settings so markdown generation also works on sites protected with HTTP Basic Auth (e.g. staging). The credentials are sent with the plugin's internal page requests.
• Show a warning on all LLMify control panel pages and in the LLMify Settings field while the plugin is disabled, so editors can see that no markdown is being generated or served.

Fixed

• The plugin settings are now shown read-only when allowAdminChanges is disabled, instead of an empty settings page. The Settings link in the LLMify navigation is also available in that case.
• Markdown generation now works on offline sites (isSystemLive: false). The plugin's internal page requests carry a Craft site token, which Craft accepts while the system is offline.
• Internal page requests that do not return a 200 are now logged as warnings instead of being skipped silently.

Added

• Add Enable llms.txt and Enable llms-full.txt site settings to disable the aggregated files per site while keeping the markdown pages available. (#24)
• Add an Include Social Links site setting with an editable Social Links table that adds a ## Social section to llms.txt and llms-full.txt. The table follows SEOmatic's "Same As URLs" until changes are saved that differ from it; social links can also be maintained without SEOmatic. (#24)

Changed

• Markdown page lookups now resolve through an indexed uri column instead of an unindexed JSON expression, speeding up .md URL serving on sites with many pages.

Added

• Add WebMCP support: the new enableWebMcp setting exposes your enabled content to in-browser AI agents (e.g. Gemini in Chrome) via the experimental WebMCP standard. A script served at /webmcp.js registers read-only search_content, get_page, list_sections, and navigate_to tools backed by the same content gating as the public markdown routes.
• Add WebMcpService::EVENT_REGISTER_TOOLS so modules and plugins can register custom WebMCP tools or modify and remove the built-in ones.
• Add mdUrl(), chatGptUrl(), and claudeUrl() Twig functions that return the markdown URL of an element and prompt links for opening it in ChatGPT or Claude.

Changed

• enableBotDetection now defaults to false for new installs. Serving markdown based on the user agent is an explicit opt-in.

Added

• Add a "Copy settings from" menu to the Site and Content settings pages to load another site's settings into the form.

Added

• Add Headless Mode setting for sites where Craft does not render the front end. Markdown is generated by fetching your front-end URLs (the site's Base URL) instead of relying on Twig rendering.
• Add API Token setting to secure rest APIs against unauthorized parties.
• Add headless API endpoints GET /actions/llmify/api/llms-txt and GET /actions/llmify/api/llms-full-txt (with a site parameter) so a headless front end can pull and re-serve the generated files.
• Add GET /actions/llmify/api/page?uri=<uri> returning a single page's stored markdown (including front matter) for a given site.
• Add POST /actions/llmify/api/convert to convert a front-end URL to markdown on demand.
• Add RequestService::fetchAndConvert() and MarkdownService::convertHtml().

Added

• Add EVENT_LLM_REQUEST to allow Insights to collect data.
• Add getMarkdownPath() helper
• Add LlmRequestType enum

Improved

• Change description db columns from string to text
• Change field description of bot settings to be more descriptive

Added

• Sections can now have a custom order in the llms.txt and llms-full.txt via Content Settings.
• Add template support for all plain text fields in site, content and llm field settings.

Improved

• Improve AI bot detection by using https://raw.githubusercontent.com/ai-robots-txt/ai.robots.txt/main/robots.json.
• Removes prerequirements in AI skill.

Fixed

• Sections with no URLs won't show up anymore in the content settings and dashboard.
• Fix 500 error if LLMify tries to generate markdown for an element without a URL.
• Disable "Add a field" button when override front matter is false.

Added

• Add dashboard with site setup scores and section-level content statistics.
• Add AI crawler detection to auto-serve markdown to known bots (GPTBot, ClaudeBot, ChatGPT-User, and more) based on user agent.
• Add <link rel="alternate" type="text/markdown"> discovery tag injection with configurable toggle.
• Add /.well-known/llms.txt route for RFC 8615 compliant discovery.
• Add per-entry exclude toggle via the LlmifySettingsField to individually exclude entries from markdown generation and all LLM outputs.
• Add SEOmatic field support for automatic front matter population.
• Add Vary: Accept header to all markdown endpoints and Vary: Accept, User-Agent to auto-serve responses.
• Add X-Robots-Tag: noindex, nofollow header to auto-serve responses.
• Add custom bot user agent configuration for extending the built-in bot list.
• Add markdown preview targets for entries and products.
• Add direct links to content and site settings from the entry sidebar and settings field.

Improved

• Reorganize plugin settings into a tabbed UI (General, Advanced) integrated into the plugin subnav.
• Enable content negotiation (autoServeMarkdown) and AI crawler detection (enableBotDetection) by default for fresh installs.
• Make markdownUrlPrefix optional, leave it empty to use ${url}.md URLs directly.
• Refactor field discovery into dedicated FieldDiscoveryService.
• Improve sidebar panel UX: show informative messages when LLMify is disabled at the site, section, or entry level instead of hiding the panel. Respect generate and clear permissions for sidebar buttons.
• Enhance markdown generation for images inside links.

Fixed

• Fix breadcrumbs in all plugin pages.
• Fix bug in field discovery service with field layout overwrites.
• Fix UI issue with front matter inheritance in the settings field.
• Fix issue with cached SEOmatic field values in llms.txt.

Fixed

• Add icon to llmify settings field.

Added

• Add Craft Commerce product support as an optional soft dependency.
• Add X-Robots-Tag: noindex, nofollow header to all markdown endpoints.
• Add Link: rel="canonical" header to individual markdown pages pointing to the original HTML page.

Improved

• Expand default remove_nodes to strip non-content elements (form, button, input, select, option, svg, script, nav, noscript).
• Remove empty markdown links left behind after image/node removal.
• Decode HTML entities in generated markdown output.

Changed

• Generalize database schema from entry-specific to element-generic (entryId → elementId, sectionId → groupId, entryMeta → elementMeta).
• Generalize all services to work with ElementInterface instead of Entry.

• Add auto serve markdown mode if the text/markdown header is present.
• Add auto serve markdown setting.
• Add blitz support for auto serve markdown mode.
• Add link to generated markdown in entry sidebar.
• Add auto try to generate markdown files before showing a 404 for requests with text/markdown and raw/${uri}.md routes.

• Add front matter support with hierarchical inheritance (Site → Section → Entry)
• Add Content Block text fields support with dot notation
• Add entry-level override settings for title, description and front matter
• Add link to section settings in entry field UI

• Fix request context issues in EVENT_AFTER_RENDER_TEMPLATE. #1

• Remove punctuation mark at the end of the llms.txt description and filter out links without title or url.

• Initial release