Version 3.8.0

We're excited to announce the release of Smartly.AI version 3.8.0, packed with new features, improvements, and fixes designed to enhance both the developer's and the end-user's experience. This update focuses on providing detailed insights, enhancing multilingual interactions, and introducing dynamic modifications for assistants, all aimed at streamlining the conversation flow and ensuring higher accuracy and user satisfaction.

🗓 Release dates

• Smartly.AI Release: 21h
• Virtual Agent Release: 27th

What’s New ?

Enhanced Conversation Module

• Detailed Interaction Logs: Every user interaction within the conversation module now includes comprehensive logs. By selecting "View JSON," users can access detailed information on dialog engine activities, NLU (Natural Language Understanding) processes (traditional, hybrid, or generative AI-driven), and custom code execution timings.
• Improved Log Visibility: We've expanded the log viewing area for better readability and clarity.
• Custom Code Processing Insights: A new "custom_code_processing_time" field in the API output now provides the execution time of custom code, offering deeper performance insights.

Advanced Hybrid NLU Capabilities

• Multilingual Support: Enhanced ability to translate user messages into the target language, improving support for multilingual interactions.
• Context-Aware Intent Detection: Improved detection of user intent over multiple messages, leveraging conversation context for greater accuracy.
• Security Enhancements: Enhanced security measures including prompt injection detection, input sanitization, and detailed logging of interaction costs and durations.

Dynamic Assistants

• On-the-Fly Custom Instructions: Users can now send custom instructions dynamically to benefit from personalized assistant responses.
• Real-Time Activation/Deactivation: Flexibility to activate or deactivate generative AI assistants instantly, adapting to user needs.
• API Enhancements: Updated /knowledge_bases and /assistants endpoints with improved documentation and the ability to manage knowledge base documents more effectively.
• Intelligent Document Handling: Optimized document search and retrieval processes, including the ability to bypass unnecessary steps and follow multiple redirections for URL-based document ingestion.

Quality of Life Improvements

• Conversation Module Fixes: Addressed an issue with the Reviews filter, ensuring smoother user experiences.
• Analytics Restoration: Restored specific analytics data from early 2023 in the Blue instance, providing users with valuable insights into their bots' performance.

This update aims to provide a seamless and secure experience for all Smartly.AI users, combining detailed analytics, enhanced multilingual support, and flexible assistant management to empower your conversational AI solutions.

Version 3.7.0

🗓 Release dates

• Smartly.AI: 15th
• Virtual Agent: 22th

What’s New ?

Changelog: Accessibility and Usability Enhancements

In our latest update, we've focused on making substantial improvements to the accessibility and usability of our platform, ensuring compatibility with RGAA guidelines. Here's a summary of the changes we've implemented:

Accessibility Improvements:

• Keyboard Navigation Enhancements: We've made significant improvements to keyboard navigation across our platform. This includes enabling the 'Escape' key to minimize the chatbot, ensuring the submit input field and button are closely linked with no space in between for a seamless visual and interactive experience, similar to the profile page setup. Additionally, the focus now correctly moves to the chatbot container upon opening, mimicking modal behavior for better accessibility.
• Semantic HTML and ARIA Updates: We've refined our HTML and ARIA usage for better screen reader support. The send button is now a proper <button> with type="submit", and messages are contained within an element with role="log". Input fields are now true <input> elements with associated labels (visible or via aria-label) to improve form accessibility. We've also added the enterkeyhint="send" attribute to optimize keyboard interaction on smartphones.
• Enhanced Focus Management: To aid users navigating with keyboards, we've ensured that focus is always visible and intuitive. The focus returns to the chatbot's reopening link after closing it with the 'Escape' key, and we've improved visibility of focus on buttons during keyboard navigation. The chatbot now opens within an iframe with a descriptive title attribute, informing screen reader users that it is a conversational tool.

Usability Updates:

• Submit Button Behavior: We've adapted the submit button's color change event to reactivate once the minimum character count is reached in the input field, rather than upon the complete deletion of the message and first character retyping.
• Modal and Popup Interactions: Closing modals with the 'Escape' key and managing pop-ups via keyboard have been streamlined for better user experience. The startup "start window" panel, icons in the webchat header, and the "display closing windows" feature are now fully keyboard-navigable.
• Focus Enhancements: The input area now correctly focuses when the chatbot opens, ensuring users can immediately start typing. We've also addressed the issue where the chatbot's focus was not always visible during keyboard navigation.

RGAA Compliance:

• Screen Reader Support: We've made adjustments to ensure our platform is more compatible with screen readers. This includes setting a proper role="log" on the message container without an unnecessary aria-label, as well as adding meaningful alt texts to all carousel and card images to describe their function or content.
• Chatbot Navigation: Enhancements have been made to enable full keyboard navigation within the chatbot and to ensure automatic reading of sent messages, aligning with RGAA standards for accessible web content.

These updates reflect our commitment to accessibility, providing a more inclusive and user-friendly experience for all our users. We appreciate your feedback as we continue to improve and adhere to accessibility standards.

Changelog: Enhancements in Generative AI and NLU Integration

Our latest updates bring significant advancements in Generative AI functionalities and Natural Language Understanding (NLU) capabilities, aiming to provide a more efficient, reliable, and user-friendly experience. Here's a detailed overview of the enhancements and new features:

Generative AI Enhancements:

• Conversations (New): We've introduced the ability to track token usage and cost for each interaction. This feature provides valuable insights into your usage patterns, helping optimize costs and efficiency.
• Assistants (New): We're excited to launch a new reranking algorithm designed to prioritize the most relevant documents at the top of your search results. Initially, we are integrating the Cohere AI Reranker, recognized for its superior performance in document relevancy ranking.
• Assistants (Improved): You now have more control over customizing the behavior of each step in your pipeline. This improvement allows for finer adjustments, ensuring that your assistants operate in perfect alignment with your specific requirements.
• Knowledge Base (New): Uploading documents to your knowledge base has never been easier, with the new capability to upload multiple documents simultaneously, enhancing efficiency and user experience.
• Knowledge Base (Fixed): The document indexation process has been significantly improved for robustness, ensuring reliability even when certain web pages are unreachable.
• Tools: We're sharing a powerful tool on GitHub designed to streamline the ingestion of numerous files and URLs at once, making it easier to manage your resources. The repository is available here.
• Models: The introduction of gpt-4-0125-preview, the latest innovation from OpenAI, is now available, offering cutting-edge performance and capabilities.

NLU Integration Updates:

• Additional NLU (Improved): Enhancing our LLM-based NLU offerings, you now have the option to select and configure your preferred Large Language Model (LLM) for enhanced natural language understanding tasks, providing greater flexibility and precision in processing natural language inputs.

Dialog API Improvements:

• In a significant update to our /dialog endpoint, we've improved error handling: now, if an error occurs during code execution on a bot, we return a more informative error code 460 instead of the generic 500. This change aims to provide clearer diagnostics and facilitate easier troubleshooting for developers.

We're excited for you to experience these enhancements and look forward to your feedback as we continue to innovate and improve.