The imports option of generateFiles() no longer splits the generated imports into one character per line. Each entry is written as one import { ... } from '...'; line.
Structured data records each table row as a Markdown table of its header row and itself, instead of a record per cell. Type tables from auto-type-table record a row for each prop instead of the TypeTable element. Re-sync your search indexes to pick them up.
remarkStructure() defaults types to table instead of tableCell.
tableRowToStructuredData() from fumadocs-core/search creates the structured data of a table row, typeTableToStructuredData() from fumadocs-typescript of a type table.
Search hooks group the rows of a table into a table item, <SearchDialogList /> renders it with Table, <SearchDialogListTable /> by default.
Algolia and Orama Cloud link to the anchor of every record, like the props of type tables.
Orama Cloud searches title, section and content by default, set params.properties to search other fields.
remarkStructure() of Sätteri records the data.structuredData of nodes in place of the nodes.
Fix images in descriptions, show the OAuth redirect URI
Images in Markdown of API pages, like the description of an operation, and in descriptions of <AutoTypeTable /> props no longer crash the page with "Element type is invalid".
The OAuth panel of the playground shows the redirect_uri of the authorization code and implicit flows, with a button to copy it: the oauthRedirectUrl route when set, otherwise the current page.
The Authorize row of OAuth schemes in the playground is one button, filling the row.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
Types longer than 500 characters in hover popups are truncated by the checker (... 26 more ...) instead of printed in full, so pages with large object types compile to much smaller MDX.
Code blocks are briefly written to the .twoslash directory while being analyzed, instead of going through filesystem callbacks, which makes the analysis about 2x faster. Add .twoslash to your .gitignore.
Identical popups of a code block are highlighted once.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
Fix images in descriptions, show the OAuth redirect URI
Images in Markdown of API pages, like the description of an operation, and in descriptions of <AutoTypeTable /> props no longer crash the page with "Element type is invalid".
The OAuth panel of the playground shows the redirect_uri of the authorization code and implicit flows, with a button to copy it: the oauthRedirectUrl route when set, otherwise the current page.
The Authorize row of OAuth schemes in the playground is one button, filling the row.
playground.authProviders handle security schemes: their rows, panels like signing in, and how values are sent. Omitted options come from the built-in providers.
Sign-ins leaving the page use useAuthRedirect().
HTTP schemes other than Basic default to their own prefix, like Token .
The username and password of HTTP Basic auth are remembered.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
Structured data records each table row as a Markdown table of its header row and itself, instead of a record per cell. Type tables from auto-type-table record a row for each prop instead of the TypeTable element. Re-sync your search indexes to pick them up.
remarkStructure() defaults types to table instead of tableCell.
tableRowToStructuredData() from fumadocs-core/search creates the structured data of a table row, typeTableToStructuredData() from fumadocs-typescript of a type table.
Search hooks group the rows of a table into a table item, <SearchDialogList /> renders it with Table, <SearchDialogListTable /> by default.
Algolia and Orama Cloud link to the anchor of every record, like the props of type tables.
Orama Cloud searches title, section and content by default, set params.properties to search other fields.
remarkStructure() of Sätteri records the data.structuredData of nodes in place of the nodes.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
remarkImage() with onError: 'hide' removes the right images when a paragraph has several, and Sätteri removes them from the Markdown too.
files code blocks of remarkMdxFiles() nest the entries under the last item of a folder (after └──), and trees indented with spaces only.
The remark and rehype plugins no longer slow down quadratically on large documents, like rehypeToc() taking 17 ms instead of 239 ms on a 5 MB page.
fumadocs-epub removes every unresolved image, the second of two adjacent ones was kept.
Sätteri: headings with expression props, like <Badge value={1} /> or local images, compile again, without these elements in their toc entry.
Sätteri: remarkStructure() records every node type listed in types, like fumadocs-core. A blockquote is one search record instead of two. In the Markdown, includes and type tables stay inside their list item or blockquote, heading IDs no longer break setext and closed ATX headings, and nested replacements no longer duplicate the source after them.
Fix images in descriptions, show the OAuth redirect URI
Images in Markdown of API pages, like the description of an operation, and in descriptions of <AutoTypeTable /> props no longer crash the page with "Element type is invalid".
The OAuth panel of the playground shows the redirect_uri of the authorization code and implicit flows, with a button to copy it: the oauthRedirectUrl route when set, otherwise the current page.
The Authorize row of OAuth schemes in the playground is one button, filling the row.
Search results no longer wrap matches in <mark>, content is the indexed Markdown. The search dialog highlights matches with the CSS Custom Highlight API, style them with ::highlight(fd-search). Custom search UIs can use useHighlightQuery() from fumadocs-core/search/client.
createContentHighlighter() is deprecated.
Removed contentWithHighlights from search results, and the renderHighlights prop of <SearchDialogListItem />.
<SearchDialogListItem /> renders the Markdown of results with the prose prose-sm typography.
Search hooks for each provider
fumadocs-core/search/client exports a hook for each search client, like useFetchSearch(). They return the props of <SearchDialog />:
data is the last successful search, its items have their content parsed into hastContent. error is set when the last search failed.
experimental_useSearch() creates a search hook from a memoized search function.
useDocsSearch() is deprecated.
useFlexsearchStatic() and useOramaCloudLegacySearch() are exported from the paths of their clients.
<SearchDialogList /> shows defaultItems without results.
useSearchList() is removed, useSearch() returns getActive(), setActive() and subscribeActive() for the active item.
<SearchDialogListItem /> renders a div instead of a button, table rows render their cells only.
renderMarkdown of <SearchDialogListItem /> renders content instead of hastContent. Without it, the string content of custom items is shown as text.
The templates of Fumadocs CLI and Create Fumadocs App use the new hooks.
Index tables by row
Structured data records each table row as a Markdown table of its header row and itself, instead of a record per cell. Type tables from auto-type-table record a row for each prop instead of the TypeTable element. Re-sync your search indexes to pick them up.
remarkStructure() defaults types to table instead of tableCell.
tableRowToStructuredData() from fumadocs-core/search creates the structured data of a table row, typeTableToStructuredData() from fumadocs-typescript of a type table.
Search hooks group the rows of a table into a table item, <SearchDialogList /> renders it with Table, <SearchDialogListTable /> by default.
Algolia and Orama Cloud link to the anchor of every record, like the props of type tables.
Orama Cloud searches title, section and content by default, set params.properties to search other fields.
remarkStructure() of Sätteri records the data.structuredData of nodes in place of the nodes.
Show page tree roots on fallback pages
On pages from the fallback page tree, layout tabs list the root folders of the main page tree again.
Restore sync index() and indexNode() in llms()
Since 16.15.9, index() and indexNode() of llms() returned a promise, so code using them as strings, like joining indexNode() results, printed [object Promise] without a type error. They return a string again when llms() receives a loader, and only return a promise when it receives a function like getSource of runtime content sources. Awaiting them keeps working in both cases.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
remarkImage() with onError: 'hide' removes the right images when a paragraph has several, and Sätteri removes them from the Markdown too.
files code blocks of remarkMdxFiles() nest the entries under the last item of a folder (after └──), and trees indented with spaces only.
The remark and rehype plugins no longer slow down quadratically on large documents, like rehypeToc() taking 17 ms instead of 239 ms on a 5 MB page.
fumadocs-epub removes every unresolved image, the second of two adjacent ones was kept.
Sätteri: headings with expression props, like <Badge value={1} /> or local images, compile again, without these elements in their toc entry.
Sätteri: remarkStructure() records every node type listed in types, like fumadocs-core. A blockquote is one search record instead of two. In the Markdown, includes and type tables stay inside their list item or blockquote, heading IDs no longer break setext and closed ATX headings, and nested replacements no longer duplicate the source after them.
Sync Mixedbread stores with sync()
sync() and toDocuments() from fumadocs-core/search/mixedbread upload a file for each page, whose chunks are its title, headings and paragraphs. Results link to their headings and render tables like other search integrations.
Stores synced with mxbai store sync need a re-sync, the frontmatter of pages is no longer read.
Tag filters match the tags of pages, locale filters results by language, and the limit of requests can only lower topK.
The Mixedbread template of Fumadocs CLI syncs a pre-rendered static.json with sync(), debounces searches by 300 ms, and reads the API key from MXBAI_API_KEY.
Ignore unsupported locale cookies
With hideLocale: 'always', createI18nMiddleware() ignores locale cookies that aren't in languages.
Limit results of the advanced search server
Without a limit in the request, it returns up to 60 results instead of every matched record.
The limit in its search option applies to requests without one.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
remarkImage() with onError: 'hide' removes the right images when a paragraph has several, and Sätteri removes them from the Markdown too.
files code blocks of remarkMdxFiles() nest the entries under the last item of a folder (after └──), and trees indented with spaces only.
The remark and rehype plugins no longer slow down quadratically on large documents, like rehypeToc() taking 17 ms instead of 239 ms on a 5 MB page.
fumadocs-epub removes every unresolved image, the second of two adjacent ones was kept.
Sätteri: headings with expression props, like <Badge value={1} /> or local images, compile again, without these elements in their toc entry.
Sätteri: remarkStructure() records every node type listed in types, like fumadocs-core. A blockquote is one search record instead of two. In the Markdown, includes and type tables stay inside their list item or blockquote, heading IDs no longer break setext and closed ATX headings, and nested replacements no longer duplicate the source after them.
Markdown and search records from the authored source
remarkLLMs() and remarkStructure() of fumadocs-core slice their Markdown from the authored source like Sätteri, instead of stringifying the syntax tree. It is much faster, Markdown is no longer escaped or reformatted, and local images keep their addresses. Re-sync your search indexes to pick up the new records.
Plugins record the Markdown of content they replace with replaceSource() or embedSource() from fumadocs-core/mdx-plugins/stringifier, in place of data._stringify, and add search records with data.structuredData. Nodes they insert otherwise are left out. A namespace limits an edit to the stringifiers of that namespace, like search for search records.
Search records are always Markdown, also in Sätteri. filterElement replaces stringify.filterElement, it chooses the JSX elements kept as HTML tags (File, TypeTable, Callout and Card by default) that search dialogs render as components. Other elements and links are replaced by their content, images are removed, and an element on one line is recorded like a paragraph.
Removed defaultStringifier() (use createStringifier()), StringifyOptions, allowedMdxAttributes, filterMdxAttributes, placeholder() (use mdxAsPlaceholder), and the stringify and mdast-util-to-markdown options of both plugins.
Included content goes through the plugins like the document's own, Sätteri includes .md files as Markdown.
<auto-files> shows its files in a code block, auto-type-table its props in a Markdown table, created by typeTableToMarkdown() from fumadocs-typescript.
remarkShow() and fileGenerator() of fumadocs-docgen record their Markdown, generators receive the file to do the same.
Records of fumadocs-obsidian resolve wikilinks, and leave comments, block IDs, callout markers and embeds out. Comments are removed before parsing, so they no longer split a paragraph.
Pages of fumadocs-python compile from their MDX, so they have Markdown and search records.
fumadocs-mdx, fumadocs-docgen and fumadocs-obsidian require fumadocs-core 16.17.0.
data is the last successful search, its items have their content parsed into hastContent. error is set when the last search failed.
experimental_useSearch() creates a search hook from a memoized search function.
useDocsSearch() is deprecated.
useFlexsearchStatic() and useOramaCloudLegacySearch() are exported from the paths of their clients.
<SearchDialogList /> shows defaultItems without results.
useSearchList() is removed, useSearch() returns getActive(), setActive() and subscribeActive() for the active item.
<SearchDialogListItem /> renders a div instead of a button, table rows render their cells only.
renderMarkdown of <SearchDialogListItem /> renders content instead of hastContent. Without it, the string content of custom items is shown as text.
The templates of Fumadocs CLI and Create Fumadocs App use the new hooks.
Fix images in descriptions, show the OAuth redirect URI
Images in Markdown of API pages, like the description of an operation, and in descriptions of <AutoTypeTable /> props no longer crash the page with "Element type is invalid".
The OAuth panel of the playground shows the redirect_uri of the authorization code and implicit flows, with a button to copy it: the oauthRedirectUrl route when set, otherwise the current page.
The Authorize row of OAuth schemes in the playground is one button, filling the row.
data is the last successful search, its items have their content parsed into hastContent. error is set when the last search failed.
experimental_useSearch() creates a search hook from a memoized search function.
useDocsSearch() is deprecated.
useFlexsearchStatic() and useOramaCloudLegacySearch() are exported from the paths of their clients.
<SearchDialogList /> shows defaultItems without results.
useSearchList() is removed, useSearch() returns getActive(), setActive() and subscribeActive() for the active item.
<SearchDialogListItem /> renders a div instead of a button, table rows render their cells only.
renderMarkdown of <SearchDialogListItem /> renders content instead of hastContent. Without it, the string content of custom items is shown as text.
The templates of Fumadocs CLI and Create Fumadocs App use the new hooks.
Sync Mixedbread stores with sync()
sync() and toDocuments() from fumadocs-core/search/mixedbread upload a file for each page, whose chunks are its title, headings and paragraphs. Results link to their headings and render tables like other search integrations.
Stores synced with mxbai store sync need a re-sync, the frontmatter of pages is no longer read.
Tag filters match the tags of pages, locale filters results by language, and the limit of requests can only lower topK.
The Mixedbread template of Fumadocs CLI syncs a pre-rendered static.json with sync(), debounces searches by 300 ms, and reads the API key from MXBAI_API_KEY.
Fix images in descriptions, show the OAuth redirect URI
Images in Markdown of API pages, like the description of an operation, and in descriptions of <AutoTypeTable /> props no longer crash the page with "Element type is invalid".
The OAuth panel of the playground shows the redirect_uri of the authorization code and implicit flows, with a button to copy it: the oauthRedirectUrl route when set, otherwise the current page.
The Authorize row of OAuth schemes in the playground is one button, filling the row.
The aiChat panel turns visible as soon as it opens, so its input can take focus right away. It still hides only after the closing transition.
Collapse the sidebar with Ctrl + B
On desktop, press Ctrl + B (⌘ + B on macOS) to collapse or expand the sidebar, in layouts that support collapsing it. Like the theme hotkey, it is ignored while typing in an editable element, or when a dialog is opened.
Fix Spacious layout on Firefox mobile
When scrolling in Firefox, the top edge of the page panel moved away from the navbar and table of contents bar, ending up in the middle of the page or off screen. It now stays below the bars, without relying on CSS anchor positioning.
Ignore unknown values in tab groups
Tabs with a groupId, including code block tabs with tab-group, applied persisted or shared values even when none of their tabs matched, for example after renaming a tab, or when tabs in the same group have different items. They then showed no content until a tab was clicked. Such values are now ignored.
New package: @fumadocs/ai-chat
The Ask AI chat for AI SDK: pass the result of useChat() to AIChatProvider, then render AIChatPanel in the aiChat option of docs layouts. Questions rest at the top while answers stream below, and Markdown is rendered block by block, so unclosed syntax never shows.
The ai/* components of the CLI install its source along with the integration, and the generated layout uses AIChat, AIChatPanel, AIChatTrigger and useAIChat.
New package: @fumadocs/ai-chat
The Ask AI chat for AI SDK: pass the result of useChat() to AIChatProvider, then render AIChatPanel in the aiChat option of docs layouts. Questions rest at the top while answers stream below, and Markdown is rendered block by block, so unclosed syntax never shows.
The ai/* components of the CLI install its source along with the integration, and the generated layout uses AIChat, AIChatPanel, AIChatTrigger and useAIChat.