You can now generate library documentation from local source code using the --local flag. Instead of pushing to a remote repository and parsing on Fern’s servers, fern docs md generate --local parses the source locally. This doesn’t require you to be logged in to Fern.
To use local generation, set your library’s input to a path instead of a git URL:
Fern no longer generates llms-full.txt, and the llms-full-txt configuration in docs.yml is deprecated. Concatenating an entire site into one file exceeded most model context windows, added heavy serving overhead, and saw little use next to llms.txt combined with per-page Markdown.
Use llms.txt to discover page URLs and fetch individual pages via their .md URLs.
You can now install the fern-docs agent skill to teach coding agents like Claude Code, Cursor, and Copilot how to work with Fern Docs. The skill covers docs.yml configuration, navigation, MDX pages, components, and changelog entries, with task-specific references that agents load on demand.
Fern docs sites can now serve author-supplied Agent Skills at the standard /.well-known/skills/ and /.well-known/agent-skills/ endpoints. Place your skill bundle under fern/.well-known/agent-skills/ and the CLI uploads it during fern generate --docs. Users install with npx skills add https://<your-domain>.
You can now add an “Install skills” button to the page action bar via page-actions.options.skills in docs.yml. The button opens a modal showing a copyable install command, the list of available skills, and a link to the skill source. The modal fetches the site’s served well-known manifest on first open, so the skill list stays in sync with what npx skills add installs.
You can now control how search results are ranked and filtered for sites with multiple products or versions using the new settings.search object in docs.yml.
prioritize-current-product boosts results from the current product higher in the ranking without hiding other products.default-filter-by-current-product auto-selects the current product as a facet filter so only current-product results appear by default.Interactive API components now render as fenced code blocks and structured content in your site’s Markdown output. AI agents consuming your docs via .md URLs, llms.txt, or llms-full.txt receive the full request and response examples without parsing HTML, reducing content disparity between the rendered page and its Markdown representation.
This applies to the following components:
You can now check for broken links on your live documentation site directly from the Fern CLI with fern docs link check. The command scrapes every page on your published site, checks all internal and external links, and reports broken (404) and blocked (403) URLs along with the source pages where they appear.
You can now serve your own robots.txt at the root of your documentation site by pointing agents.robots-txt at a file in your repo. Use this to opt in or out of specific AI crawlers like GPTBot or ClaudeBot, gate sensitive sections from indexing, or signal training and search preferences with the Cloudflare Content Signals Policy. Your file is served verbatim at /robots.txt, and Fern appends a managed block disallowing internal API routes after your content.
Multi-source docs lets each team publish independently to a shared custom domain — for example, docs.nvidia.com spans sub-paths like /nvcf, /brev, and /aiperf, each owned by a different product team and repository.
Set multi-source: true on the instance and reference a global theme for consistent branding across repositories.
AI 代理、MCP 客户端和 API 目录爬虫现在可以自动发现您的 API。每个 Fern Docs 站点都会暴露一个基于标准的（RFC 9727）/.well-known/api-catalog 端点，该端点从您可见的 API 参考导航生成。
您现在可以将 <Prompt> 操作指向任何 AI 工具，而不仅仅是 Cursor、Claude 和 ChatGPT。将读者引导到 Perplexity、您自己的内部 AI 助手或任何其他 URL。
从我的 OpenAPI 规范生成 TypeScript SDK。请遵循 [TypeScript SDK 快速开始](https://buildwithfern.com/learn/sdks/generators/typescript/quickstart.md)。GitHub Actions 预览工作流 现在有一个可选的配套 cleanup-preview.yml 工作流，可以在 PR 合并后删除拉取请求的预览部署，这样过期的预览就不会留存。
要采用此工作流，请在现有预览工作流旁边添加 .github/workflows/cleanup-preview.yml。
更新的 GitLab CI/CD 流水线现在包含一个 cleanup_preview 阶段，在合并请求合并到默认分支后删除该 MR 的预览部署，避免陈旧的预览残留。
要采用此工作流程，请用更新的版本替换您的 .gitlab-ci.yml 文件。
使用 guide 和 overview 布局的页面现在可以在移动端和平板设备视口中在头部下方显示粘性目录栏。该栏显示滚动进度指示器和当前标题，点击时展开为完整目录。
在 docs.yml 的 layout 下添加 mobile-toc: true 来启用：