Released: Sep 5, 2024

V3.0 中的更新

特性

Retrieval-Augmented Generation (RAG)

• Retrieve Project Information to Provide More Context to the AI - When using the AI Positron Assistant add-on with Oxygen versions 26.1 or newer, certain actions (New DITA Topic, Add Structured Content, Generate Documentation Draft) now leverage the Retrieval-Augmented Generation (RAG) process. This output optimization process enhances the AI's responses by using information from the current project opened in the Project view to enrich the AI context for you to receive more meaningful and project-targeted responses.
• Function Calls - Within the definition of custom actions, you can reference existing functions that are called by the AI engine to interact with the application. This gives actions more context information and allows for more accurate answers to be constructed by the AI. The current available reference values for pre-defined function calls are:
  • get_related_content_from_project - Retrieves content from the user's local project based on given key words. It is limited to return a maximum of 50k characters and works with Oxygen version 26.1 or newer.
  • get_related_resources_overview_from_project and get_content_for_document_id - Retrieves an overview of a maximum of 5 documents that each contain an ID along with the most relevant information from each document (usually titles, key words, short descriptions). The second function can be called by the AI to retrieve the content of a certain document. It is also limited to return a maximum of 50k characters and works with Oxygen version 26.1 or newer.
  • get_current_document_plain_text_content - Obtains all text without markup from the current document open in the editor.
  • get_current_document_marked_up_content - Obtains all text with markup from the current document open in the editor.
  • get_content_around_caret - Obtains size-limited content around the current cursor location within the document open in the editor.

Interface Improvements

• AI Positron Assistant Widget - For documents that are open in the Author visual editing mode, there is now an AI Positron Assistant drop-down widget in the top-right corner of the editor. It offers a very quick and convenient way of accessing useful AI actions to create and improve the structure and content in the current document.
• In-place AI Prompt Input Field - The AI Positron Assistant drop-down widget contains a helpful Rewrite content action that opens a floating input box close to the text cursor location where you can provide the AI with instructions on how to rewrite the selected content (or the current paragraph if there is no current selection).
• Regenerate Response - After you receive a response from the AI, you now have the option to ask the AI to regenerate another answer. With OpenAI, you can also decide which engine model to employ when regenerating the response.
• Preview Responses When Using AI Features in Text Mode - When using AI actions while working on your documents in the Text mode, you now have the ability to preview the AI-generated response before inserting the changes in the document.
• Access Quick Assist Actions From Vertical Stripe Icon - Custom user-defined AI actions can now be configured to appear as Quick Assist fixes in the editor. This feature offers you convenience by providing direct access to frequently used fixes without the need to navigate through menus, and supports customization to better suit your workflow and preferences.
• Easily Choose Between AI Models - The AI model used for each particular action or chat session can easily be changed from the convenient drop-down list located in the toolbar of the AI Positron Assistant view. The default model is now GPT-4o, which is an advanced model that provides better results but is more expensive than other models, so it is quite helpful to have a very quick and easy way of changing it for actions that may not require advanced results. The chat panel now also specifies which AI model was used for each response.

AI Models

• GPT-4o-mini AI Model - The GPT-4o-mini AI model has been added to the list of available models. You can choose to use this new model on a case-by-case basis or you can configure the connection settings to make it the default model used for all actions.
• Improved Credit Usage Efficiency - The credit system has been updated to enhance cost efficiency. Now, input tokens and output tokens are treated differently when calculating consumed credits, with input tokens being less expensive than output tokens.

AI-Powered Actions

• Proofread Action - When editing DITA XML documents, a new 'Proofread' AI action is now available (in the Review category) to help you identify potential issues in your content. The AI will proceed to add comments in the content regarding logical consistency problems, grammar or spelling mistakes, or readability and comprehension issues.
• Resolve Comments Action - A new AI action titled 'Resolve Comments' has been added in the Review category that enables the AI to change the selected content based on the suggestions found within comments. It then also removes the comments, saving you time in the cleanup process. The action becomes available when working with DITA XML documents.
• Improve Structure Action - A new 'Improve Structure' AI action (in the Rewrite category) can now be used when working with DITA XML documents to instruct the AI to improve the selected DITA XML content by adding additional structure or inline elements.
• Add Structured Content Action - In the Content Generation category, a new Add Structured Content AI action has been implemented that enables you to continue the content from a DITA XML document with additional structured content generated based upon similar content from the current project (if available). This helps the AI with context and leads to more accurate and higher quality results.

Web Author

• Edit Messages Sent to AI - You can now edit your message that you send to the AI by clicking a convenient edit button that appears to the right of your message in the response area. Once you submit your edited message, the AI regenerates the response. This streamlines the process of fine-tuning the results to suit your needs.
• Selected Content Sent to AI for Context - When directly initiating a new chat with the AI (without invoking an action), any content selected in the main editor area is now automatically passed as context for the conversation. This feature allows you to effortlessly process content within the editor.
• Improve Structure Action - A new 'Improve Structure' AI action can now be used to instruct the AI to improve the selected DITA XML content by adding additional structure or inline elements.
• Proofread Action - When editing DITA XML documents in Oxygen XML Web Author, a new Proofread AI action is now available to aid you in identifying potential issues in your content. The AI will then proceed to add comments in the content regarding logical consistency problems, grammar or spelling mistakes, or readability and comprehension issues.
• Resolve Comments Action - A new Resolve Comments AI action has been implemented in Oxygen XML Web Author as well, enabling the AI to modify selected content based on suggestions in the comments. It also automatically removes the comments, saving you time during the cleanup process. This action is available when working with DITA XML documents.
• Generate Image Alternate Text AI Action - When editing DITA XML files, you can now use the new 'Generate Image Alternate Text' AI action to generate an alternate text for an image that is selected in the editing area.
• Function Calls - Within the definition of custom actions, you can reference existing functions that are called by the AI engine to interact with the application. This gives actions more context information and allows for more accurate answers to be constructed by the AI. The current available reference values for pre-defined function calls for Web Author are:
  • get_current_document_plain_text_content - Obtains all text without markup from the current document open in the editor.
  • get_current_document_marked_up_content - Obtains all text with markup from the current document open in the editor.
  • get_text_around_caret - Obtains size-limited content around the current cursor location within the document open in the editor.