CKEditor 5 changelog

This is the CKEditor 5 changelog guide. Here you will find information about the most important changes introduced in the release, new features, and bug fixes.

Information about both major and minor breaking changes is available, too, if the release introduces them. You can read more about breaking changes in CKEditor 5 in the Versioning policy guide.

You can find more information about each release in the blog posts linked at the start of each entry.

This guide provides the changelog information for the 3 latest releases of CKEditor 5. For older releases, refer to the release notes on GitHub.

We are happy to announce the release of CKEditor 5 v48.4.0.

The new Context Library grounds CKEditor AI in your organization’s knowledge instead of generic instructions. A context is a named container managed on the AI service that holds reusable prompts and reference files, such as a style guide, a glossary, or compliance rules. Once attached, every AI Chat conversation, Quick Action, Review, and Translate run follows them.

Contexts can be applied environment-wide by an administrator, per editor instance via the new config.ai.defaultContext option, or per call in programmatic flows, and they stay invisible to the end user. Learn more in the documentation.

CKEditor AI now knows your editor better. AI Chat requests include a compact snapshot of the loaded features and their configuration, so the responses stay within what your editor supports and respect your configured values, like the allowed heading levels, font sizes, or color palettes. The result: AI changes that apply cleanly to your exact setup.

Feature understanding works out of the box, with no configuration required, and this is only the first iteration. Support for custom plugins, dynamic data features, and working with comments and suggestions is planned. See the documentation for details.

CKEditor AI now analyzes images embedded in the document, so it can describe them, generate captions, or take their contents into account when editing the surrounding text. It works in AI Chat and document processing. Learn more in the image analysis section of the integration guide.

The new config.trackChanges.clipboardMode option controls how Track Changes suggestions behave when content is copied or cut. The default 'keep' places the selected content on the clipboard as-is, while 'accept' resolves the suggestions in the copy, so the clipboard holds the final text, as if the suggestions were already accepted. The source document stays untouched, which makes this especially useful when pasting content outside the editor.

This release brings a set of upgrades to the Tables feature, focused on everyday editing:

• Split multiple cells at once: select several cells and split them all horizontally or vertically in one go, instead of repeating the action cell by cell.
• Horizontal scrolling for wide tables: tables wider than the editor now scroll horizontally instead of overflowing or squeezing the page layout, thanks to the new TableScroll plugin.
• Pixel-based column widths: the Column resize feature now supports widths in pixels, following the table width unit, and an exact column width can be set through the cell width field. Widths set via the resize handle and the properties form now stay in sync.

Inserting a block widget, such as an image or a table, no longer drops the active text formatting. The editor now carries selection attributes like bold, italic, or font styles over the widget, so typing after it (or inside a newly inserted table) picks up right where you left off.

• Added support for inline roots across the AI features, including AI Chat, Quick Actions, Review, and Translate.
• Programmatic AI actions now support multi-root editors and multi-editor Context setups, and processDocument() accepts new capabilities options that enable reasoning and web search.
• In multi-editor setups, AI Chat now re-uploads only the documents that changed, instead of every document on any change.
• The AI suggestion status indicators (Accepted, Rejected, Outdated) have a refreshed look, and the Outdated indicator now shows a reason-specific tooltip explaining why the suggestion became stale.
• AI interactions that modify content are now disabled when the target editor is read-only. In setups with multiple editors this is decided per editor, so interactions targeting editable editors stay available.
• Inline root corrections across features: Find and replace now finds matches inside inline roots, while Show blocks and Footnotes no longer incorrectly work with them.
• Link improvements: decorators can now be configured while creating a link, before it is inserted, and the “Link properties” button is no longer disabled for empty links when config.link.allowCreatingEmptyLinks is enabled.

• ai: Added support for multi-root editors and multi-editor Context setups to the AI Translate API (AITranslateGateway).
• ai: Added support for multi-root editors and multi-editor Context setups to the AI Review API (AIReviewGateway).
• ai: Added support for multi-root editors and multi-editor Context setups to AIDocumentProcessingGateway#processDocument(). Replaced the root option with roots and changed the return type from AIDocumentProcessingRunResult to AIRunResult<AIDocumentProcessingRunResult>.

• ai, collaboration-core: Added an optional root option to the Document Processing API’s processDocument() method, allowing integrators to target a specific root in multi-root editors and multi-editor Context setups.

• ai: Updated the appearance of the AI suggestion status indicators (Accepted, Rejected, and Outdated). Added a reason-specific tooltip to the Outdated indicator that explains why the suggestion became outdated.

• ai: Added optional capabilities to the AI Document Processing API’s processDocument() options, allowing callers to enable reasoning and web search for AI requests.

• ai: Added a snapshot of the editor’s loaded features and their configuration to AI Chat requests, allowing the agent to tailor its responses to the editor’s capabilities.

• ai: Added support for inline roots across the AI features, including AI Chat, Quick Actions, Review, and Translate.

• ai: Disabled content-modifying AI interactions, such as applying or inserting a suggestion, when the target editor is read-only. In setups with multiple editors, interactions targeting editable editors remain available.

• ai: Added a new configuration option, config.ai.extraHttpHeaders, allowing the AI service to analyze document images hosted behind authentication, for example from a private CDN.

• ai: Introduced the ai.defaultContext configuration option, allowing integrators to attach administrator-managed contexts from the Context Library, such as reusable prompts and reference files, to AI requests without exposing them in the user interface. This configuration is supported across all AI features, including AI Chat, AI Quick Actions, AI Review Mode, and AI Translate.

• link: Added support for configuring link decorators while creating a link, before inserting it into the document. Closes #20201.

• table: Added horizontal scrolling for tables wider than the editor, preventing them from overflowing or squeezing the page layout. This behavior is provided by the new TableScroll plugin.

• table: Preserved active text formatting, such as bold, italic, and font color, when inserting a table. Typing in any cell of the new table now continues the formatting used before the table. Closes #17152.

• table: Added support for column widths in pixels to the table column resize feature when the table uses pixel widths. The cell width field can now set an exact width for an entire column in a resized table. Closes #14236.

• table: Added support for splitting multiple table cells at once. Selecting several cells and choosing “Split cell vertically” or “Split cell horizontally” now splits each selected cell in one operation.

• track-changes: Added the trackChanges.clipboardMode configuration option to control whether suggestions are preserved or accepted when content is copied or cut. The source document remains unchanged.

  The default 'keep' mode places the selected content on the clipboard with its suggestions, while 'accept' places the final content as if the suggestions were accepted.

• widget: Preserved active text formatting when inserting a new paragraph after a block widget. The editor now copies selection attributes, such as bold, italic, and font styles, from the text preceding the widget. Closes #17152.

• ai: Fixed an editor crash that occurred when applying an AI Chat change in specific multi-level list scenarios.

• ai: Preserved comments and suggestions when applying an AI change to the surrounding content.

• ai: Disabled dragging and dropping files or URLs into the AI Chat panel while an AI response is being generated. Previously, dropped resources were uploaded and then discarded when the response finished or was interrupted.

• ai: Disabled the “Ask AI” quick action button while an AI Chat response is being processed, preventing a new chat interaction from starting before the current one finishes.

• ai: Fixed an editor crash that occurred for some AI Chat queries when General HTML Support was enabled and the content included an <h1> element.

• ai: Updated the AI Chat feed to scroll to every new error message.

• ai: Disabled AI Chat quick actions while a conversation is loading from history or a reply is streaming. Previously, they could be triggered before the chat was ready.

• ai: Fixed an issue where the AI Chat feed sometimes appeared stuck after submitting a message while a large document was loaded.

• ai: Fixed the AI Chat submit button remaining disabled after starting a new chat while a file or URL was still uploading in a previous conversation. Starting a new conversation now resets the upload progress state.

• ai: Fixed the formatBlock suggestion marker being cut off in AI suggestion previews, including the AI Chat feed.

• engine: Fixed a memory leak that occurred when an editor instance remained referenced after being destroyed.

• find-and-replace: Fixed an issue where Find and Replace did not find matches inside inline roots.

• find-and-replace: Fixed a memory leak that could degrade editor performance during long editing sessions after replacing all occurrences of a search phrase.

• footnotes: Prevented footnotes from being inserted into inline roots. Footnotes within inline roots now use the first non-inline root for the footnote definitions container instead of omitting it.

• fullscreen: Prevented contextual balloons from overlapping the main editor toolbar in fullscreen mode when scrolling through elements taller than the visible editor area, such as large tables. Closes #20194.

• link: Fixed the “Link properties” button being disabled for links with an empty URL, even when config.link.allowCreatingEmptyLinks was enabled.

• mention: Prevented the mention suggestions panel from overflowing the viewport on narrow screens. When the caret is close to the screen edge, the panel now shifts horizontally to keep the entire list visible. Closes #20182.

  Thanks to @ELHart05.

• show-blocks: Prevented Show Blocks from applying to inline roots. The toolbar button is now disabled when the editor has no block roots.

• style: Prevented style previews from overflowing their buttons in the Styles dropdown. Closes #20200.

• table: Fixed a crash (conversion-slot-filter-incomplete error) that occurred when loading or pasting content containing multiple <table> elements wrapped in a single aligning <div>. Closes #20209.

• ui: Fixed the contextual balloon closing unexpectedly when using its “Previous” and “Next” navigation buttons. Clicking these buttons no longer moves focus out of the editor, so focus-sensitive views such as the balloon toolbar stay visible.

• ai: Exposed structured backend error data from AI gateway connectors, such as an issues list describing fields that failed validation, under result.error.data.backendData.
• ai: Updated AI Chat in multi-editor setups to re-upload only changed documents. Previously, changing one document re-uploaded documents from all editors.
• document-outline: Added support for recognizing all heading elements (<h1> through <h6>) in Document Outline and Table of Contents, regardless of whether Heading or General HTML Support handles them. These elements now receive an id attribute in the document data and a headingId attribute in the model.
• Separated editor UI styles from content styles. No visual or functional changes are expected, but integrations with heavily customized styling should verify their appearance after updating.

We are happy to announce the release of CKEditor 5 v48.3.1.

The release addresses vulnerabilities identified in the protobuf.js package, used within our @ckeditor/ckeditor5-operations-compressor package for real-time collaboration.

Our analysis confirms that vulnerabilities do not affect CKEditor 5.

This release primarily aims to ensure that our customers using real-time collaboration features do not encounter unnecessary security alerts from their scanning tools. We are committed to maintaining the highest security standards, and this update reflects our ongoing efforts to safeguard user environments proactively.

We are happy to announce the release of CKEditor 5 v48.3.0.

Until now, using CKEditor AI meant mainly going through its built-in UI. This release extends the programmatic APIs and opens the door to more custom AI workflows. Integrators can trigger AI from their buttons, process documents automatically in the background, or run AI server-side with no editor interface at all using the Server-side Editor API.

• AI Document Processing: Run any custom, document-level prompt entirely from code with no UI involved, for automated jobs like summarizing, reformatting, or enriching content in the background.
• AI Review: Trigger built-in or custom review commands, such as proofreading, clarity, or tone, from code so you can build automated quality gates into your editing workflow.
• AI Translate: Translate a document into a target language on demand, with or without the translation UI.

See the Using CKEditor AI programmatically guide for details.

When AI and people edit the same document, reviewers need to know who proposed what. AI suggestions can now be visually marked as AI-generated, so teams can give machine-proposed changes the right level of scrutiny, keep a clear audit trail of where content came from, and meet editorial or governance policies that require disclosing AI involvement.

The feature is opt-in, and you can choose between a pill view or AI author view. Read more in the Marking AI-generated suggestions guide.

Editors that split content into separate areas, such as email layouts, structured documents, or CMS templates with distinct regions, can now use CKEditor AI with full production confidence.

We promoted multi-root and multi-instance support to stable so AI Chat and Review consistently read context from and act on the correct region. Adding or removing editor instances at runtime, including the empty “no editors” state, is handled robustly, keeping every AI request scoped to the root the user is working in.

• Images in inline roots. Images are now supported in inline roots. A block image that cannot be placed at a given position, for example when pasting, dropping, or loading data into an inline root, now degrades to an inline image instead of being dropped. The image type, caption, and style controls adapt to what the position allows.
• Keyboard-accessible media embed resizing. Media embeds now include a keyboard-accessible resize UI: a toolbar dropdown and standalone buttons for predefined sizes, plus a balloon-hosted input for custom widths.
• AI Chat and AI Review fixes. This release resolves a range of AI Chat and Review issues affecting both reliability and presentation, including suggestions that did not appear or apply, crashes on certain historical or marker-heavy content, and rendering glitches in Safari. Preview content is now selectable for direct copying, long text and URLs wrap cleanly in the feed, and several commands are translated in non-English interfaces.
• Paste from Office and Excel Online. Word footnotes are no longer malformed when the Footnotes plugin is enabled, pasting a list followed by a paragraph aligned to an earlier list’s margin no longer throws an error, and ranges pasted from Excel Online no longer insert the clipboard’s CSS <style> block as visible text.
• Footnotes. Fixed the first footnote reference disappearing with a starting value of 0 under numbering styles that do not support it, and aligned references with the list when using roman numbering at counter values of 4000 or above.
• Emoji. The plugin no longer blocks editor startup, resulting in noticeably faster load times, and multiple editors sharing the same definitionsUrl with different useCustomFont settings no longer interfere with each other’s emoji data.
• Accessibility. Comment thread accessible names now include the first comment’s text and announce reply counts, and AI-proposed track changes suggestions now state their AI origin in their accessible name.
• Type around buttons on touch devices. Tapping the buttons that insert a paragraph above or below a selected widget now works on Android and iOS. Previously, these taps did not insert a paragraph. Thanks to @ELHart05 for contributing this fix.

• ai: Changed the signature of AIGateway.apply(). applyMethod is now a property of the second argument (an options object) instead of a positional string: replace apply( result, 'suggest' ) with apply( result, { applyMethod: 'suggest' } ).

• ai: Tightened the return types of several AI Chat and AI Review getters and methods to ReadonlyArray / ReadonlyMap. They now return copies of the original collections to prevent accidental mutation of internal state.

  Updated methods are: AIChatContext#getPendingContextItems(), AIChatContext#getSentContextItems(), AIReviewRunResult#affectedBlocks and AIGateway#mergeChangesIntoContent().

• ai: Fixed a CSS specificity conflict that made the AI Chat balloon width depend on stylesheet import order. The AI Chat balloon now sizes to its content without conflicting with default dialog styles.

• ai: Changed the CSS selector used to set the AI Chat balloon width from .ck-ai-chat-balloon to .ck-ai-chat-balloon-main. Custom styles that set the AI Chat balloon width by targeting .ck-ai-chat-balloon may no longer take effect and should target .ck-ai-chat-balloon-main instead.

• ai, track-changes: Suggestions created using AI features can now be visually marked as AI-generated to be distinguished from manual edits. See config.trackChanges.showAISource and config.trackChanges.aiAuthor.
• ai: Introduced a programmatic API for AI Document Processing. See the Using CKEditor AI programmatically documentation for details.
• ai: All AI features now report their errors through a single pipeline. Applications can monitor AI failures across chat, chat history, actions, and review, and forward them to their own error-tracking tools.
• ai: Introduced a programmatic API for the AI Translate plugin. See the Using CKEditor AI programmatically documentation for details.
• ai: Introduced a headless programmatic API for the AI Translate plugin. See the Using CKEditor AI programmatically documentation for details.
• collaboration-core: Added the PillView UI component, which displays a pill with an icon, label, and tooltip. See the API documentation for details.
• image: Images are now supported in inline roots. A block image that cannot be placed at a given location (for example, when pasting, dropping, or loading data into an inline root) now degrades to an inline image instead of being dropped. The image type change, caption, and style controls now adapt to the allowed conversion and become unavailable when that conversion is not allowed at the current position.
• media-embed: Introduced a keyboard-accessible resize UI for media embeds: a toolbar dropdown or standalone buttons for predefined sizes and a balloon-hosted input for custom widths.

• collaboration-core, track-changes: Fixed a redundant keyboard tab stop on the “AI-generated” pill shown for AI suggestions. The AI origin it signals visually is conveyed to assistive technologies through the suggestion’s accessible name.

• ai: Pasting spreadsheet or word-processor content into the AI Chat prompt input now keeps the cell text instead of attaching the accompanying preview image.

• ai: Fixed an issue where AI-suggested insertions generated by AI Chat were not displayed in the chat feed and were not applied to the content in some scenarios.

• ai: Made AI-generated content displayed in the AI Quick Actions balloon and the AI Chat suggestion preview selectable for direct copying.

• ai: Roots added at runtime with MultiRootEditor#addRoot() were sent to the AI service without a title or description, which could cause area-scoped AI Chat requests to target the wrong root.

  Now the AI features read each editor root’s title and description from the root’s attributes if available, and fall back to the editor configuration otherwise.

• ai: Improved the HTML output of the AI Quick Actions suggestion preview. Completed previews no longer contain temporary streaming elements.

• ai: The tooltip displayed when hovering over an AI Chat context chip now wraps long URLs.

• ai: Long text and URLs displayed in the AI Chat feed now wrap and no longer overflow the message container.

• ai: Fixed an AI Chat crash when loading a historical conversation containing changes for an editor that was never created in the current session.

• ai: The conversation title animation is no longer played when opening a chat from history. The animation is now shown only when the AI generates a title for the first time.

• ai: The mini toolbar in the AI Actions dialog no longer overlaps the vertical scrollbar.

• ai: The AI Review programmatic API no longer switches the editor to the review tab when a review run fails validation (for example, an unknown command or a missing model). A failed call now leaves the previously active tab in place.

• ai: Fixed an editor crash that occurred when AI Chat processed responses for document content that included comment or suggestion markers. This was most often reproducible with the General HTML Support plugin enabled.

• ai: The balloons in the AI Chat context chips row are now rendered above the AI overlay backdrop instead of behind it.

• ai: Fixed an issue where toggling AI Review or AI Translate tabs would reset their state.

• ai: Switching tabs while the AI plugin is toggled no longer duplicates views.

• ai: The AI Review mode “Adjust length” and “Adjust tone and style” commands, along with their dropdown options, are now translated when a UI language other than English is used.

• ai: AI Chat and Review suggestions now render in Safari when the AI panel is toggled while the request is being processed or after quickly switching between AI panels.

• collaboration-core: Removed the misplaced affectsData property from the CollaborationOperation interface. The property is specific to MarkerOperation. Cast to MarkerCollaborationOperation to access it.

• comments: The accessible name of a comment thread now includes the text of its first comment. Single-reply threads now announce the reply count instead of repeating the author name.

• core: Replaced a cryptic multi-root-editor-root-initial-data-mismatch error thrown when config.roots is an object with a custom prototype or a class instance. The editor now throws a dedicated editor-create-roots-not-plain-object error with a clear message.

• core: An invalid tag name passed to config.root.element or config.roots.<rootName>.element now throws a clear CKEditorError with code editor-wrong-element-name instead of a cryptic InvalidCharacterError from the browser renderer.

• emoji: Reduced editor startup time by preventing the emoji plugin from blocking editor initialization.

• emoji: Fixed an issue where multiple editor instances on the same page could interfere with each other’s emoji data when they shared the same definitionsUrl but used different useCustomFont settings. Depending on which editor initialized first, some editors could display a restricted emoji list when a full one was expected, or the other way around.

• engine: Fixed incorrect ordering of markerToElement boundary elements when multiple markers share the same end position. The closing elements are now inserted in reverse opening order to preserve nesting in the output. Closes #20173.

• footnotes: Fixed the first footnote reference disappearing when the footnote list’s starting value is set to 0 and the active numbering style, for example lower-alpha or lower-roman, does not support that value.

• footnotes: Fixed a mismatch between footnote references and the footnote list when using lower-roman or upper-roman numbering with counter values of 4000 or above. Both now consistently fall back to decimal numbering, as required by the CSS counter style specification.

• paste-from-office: Pasting content from MS Office with footnotes no longer results in malformed footnotes when the Footnotes plugin is enabled.

• paste-from-office: Pasting content from MS Office no longer throws an error when a list is followed by a paragraph aligned to the margin of an earlier list.

• paste-from-office: Pasting content from Excel Online no longer inserts the clipboard’s CSS <style> block as visible text. Closes #20188.

• real-time-collaboration: In multi-editor setups using a Context mechanism, an individual document can now be flushed on the server without affecting other editors. Previously, all editors within the Context instance turned read-only. Now only the editor connected to the flushed document becomes disconnected, while the remaining editors stay connected and editable.

• table: Upcasting table content with scoped header cells into a context that does not allow tables, for example an inline editor root, no longer throws an error.

• track-changes: When a track changes suggestion comes from AI, its accessible name now states that it was proposed by AI.

• widget: The widget type around buttons now insert a paragraph when tapped on touch devices. Previously, taps on the buttons were ignored on Android and iOS, and only selected the widget. Closes #20103.

  Thanks to @ELHart05.

• core: Values provided via the root.description and root.title configuration options are now stored in the model as the $description and $title attributes of the $root element and persist through collaboration sessions. See #10327, #10285, #10333.
• track-changes: Marked Suggestion#isExternal as read-only, matching Comment#isExternal. Changing this property should not have been possible and could lead to errors.
• ui: The custom position callback in DialogDefinition now receives both the visible DOM root Rect and the general DOM root Rect, making it possible to position the dialog even when the DOM root element is off the screen or cropped by an overflowing ancestor. Added the getRootName() option to DialogDefinition to control which DOM root the dialog is positioned relative to, improving positioning in multi-root editor setups.