Release Notes

With Agentic AI, your AI agents can now independently complete tasks and make autonomous decisions. With LoyJoy, you use Agentic AI directly at the customer interface. And this is in conjunction with tools such as product consulting, web search, and of course connected to your knowledge database. We connect the world of process automation with Agentic AI for the most powerful AI agents on the market!

---

In our brand new video, Ulf shows you how to build and get your own LoyJoy AI Agent up and running from scratch in just 9 minutes.

Agentic AI – Your AI agents are getting smarter 🤖

Thanks to the new AI Agent module, your AI agents can now perform actions independently. The AI Agent module consists of instructions (“prompts”) and tools. This allows you to give your agent instructions on how to respond to specific situations and which tools to use for that purpose. The tools are APIs or other interfaces that your agent can use to retrieve information or perform actions, thus handling customer inquiries.

With our new AI Agent module, we are equipping your chatbots with the ability to solve complex tasks independently. The AI Agent module makes the old GPT modules obsolete.

The list of tools currently includes:

• Executing processes in LoyJoy,
• Knowledge search (in the LoyJoy knowledge database),
• Product search (in the LoyJoy product database) and
• Product display,
• Web search

and will be expanded in the future.

We have already published the new template Product Finder Agentic AI in the LoyJoy Template Store, which you can use directly to create your first agent with Agentic AI.

➡️ Here goes to the template

Further Optimizations 🦄

In addition, we have created the option to use Lottie-animated chat bubbles for your chatbots. This feature allows for an even more appealing and dynamic design of conversations with your customers.

Improved scrolling behavior: The scrolling behavior in the chat widget has been optimized to enhance user-friendliness. The chat no longer automatically jumps to the latest messages.

With our September release, we’re introducing a long-awaited feature: versioning. Now you can review changes to your agent in peace and publish them for your customers only when you’re ready. Customers will only see your changes after you publish them. This feature is now available to all customers. Enterprise and Unlimited plan customers can also restore previous versions and have access to the last five versions.

GPT-5 and Mistral are here: We’ve expanded the LoyJoy Platform with four powerful new AI models, all of which can be hosted in the EU.

---

Release Highlights 🚀

Versioning – Keep Track of Your Changes 🔄

Until now, changes to an agent went “live” immediately and were instantly visible to your customers. But that’s not always desirable. Thanks to the new versioning feature, you can now make changes, test them, and publish them when you’re ready. Simply click the new “Publish” button in your agent. Only then will your customers receive the new version of your agent.

The button is clickable whenever there are unpublished changes.

This feature is now available in all LoyJoy plans.

For Enterprise and Unlimited plan customers, there’s even more: Under the new ‘Versioning’ tab in the Manager, you can view a history of your changes, restore previous versions, and ensure you always have control over your projects. The last five versions are always saved.

The first entry is the version currently live, indicated by the green ‘Live’ button. All other entries are saved versions that you can view and restore. You can also copy versions and use them in a new agent. Each version has a name, creation date, the name of the user who created it, and a description of the changes. Add a comment to a version to easily identify it.

With the ‘Restore’ button, you can restore a previous version. The restored version moves to the top and becomes the active version. The previously active version is then saved as a new version below.

➡️ Click here for the guide

More Choice & Power: New AI Models in the LoyJoy Platform 🇪🇺

In addition to new agent capabilities, you now have more control over the technology powering them. We’ve expanded the LoyJoy Platform with four powerful new AI models:

• Scaleway gpt-oss-120b – the new open-source model from OpenAI!
• GPT-5 Nano
• GPT-5 Mini
• GPT-5
• Mistral Medium

All four models can be hosted directly from the EU, ensuring maximum data security and GDPR compliance. With models from Scaleway and Mistral, we’re taking it a step further by offering AI models from European providers for the first time.

We’ve also made improvements under the hood: The GPT Knowledge Module and more are now significantly more powerful. Your agents can process information even faster and more accurately.

Our July update opens up new scope for you: automatic AI testing, fresh model options, a flexible role system, deeper context integration via MCP and animated avatars - all to build faster, manage more cleanly and communicate more personally.

---

🌟 Highlights at a glance

AI test automation - save time when testing your configurations

This really saves you time! To test the AI in a LoyJoy agent, you no longer need to type your questions into the chat and wait for the answer. Instead, simply upload your up to 100 test questions as an Excel file. The AI with your configuration will answer them automatically and you can download the results for review.

You can easily test every change to your AI settings such as prompt or model: Go to your agent and on the ‘Knowledge’ tab, you will find the ‘AI Test Automation’ button, where you can then upload your Excel list. ➡️Click here for instructions

New AI models: GPT-4.1, GPT-o4-mini & GPT-o3

Choose the right model for your goal:

• GPT-4.1 is ideal for complex dialogues with a lot of context, demanding consultations or dynamic content with high quality requirements.
• GPT-4.1 mini offers a good balance of quality and speed - ideal for interactive product finders, simple advice or feedback processes where reliability is important but maximum depth is not.
• GPT-4.1 nano** is the lightest version and ideal for clearly structured, shorter enquiries - for example simple FAQs, surveys or quick introductory dialogues.
• GPT-o4-mini: With reasoning! Compact, fast and ideal for standard dialogues or supporting tasks.
• GPT-o3: The most powerful model in the O series. Particularly suitable for complex applications with high demands on precision, depth and response quality - for example for data-intensive analyses or complex decision-making processes.

New roles & rights system

A granular roles and rights concept is important for enterprises. We have made 10 new roles available. These roles can be changed at any time by owner/owner role. You can find an overview of all roles here.

MCP (Model Context Protocol): Context & tools via stream

The MCP is now also compatible with O-Series models. This allows you to give your agent in LoyJoy access to external tools or data sources. To do this, enter your MCP client in the GPT Knowledge module. It must support Streamable HTTP.

💡 What does this do for you?

• Remote control: Use your own AI agent in the browser, such as ChatGPT or Gemini, to access LoyJoy directly, e.g. to query analytics data or analyse processes.
• More context sources: Connect knowledge databases, APIs, product data, etc.

➡️To the blog article + instructions

Animated avatars for your chatbot

Would you like to bring your LoyJoy agent to life with an avatar? It’s very easy now. Just enter a few details from your Hey-Gen or D-ID account and you’re ready to go. It really only takes a few minutes!

➡️Example & screenshot: HeyGen live demo

Further improvements

• All manager translations are now supported by gpt-4o-mini for context-aware language quality.
• New model checking rule ensures images always carry ALT text - improves SEO & accessibility.
• API client module allows custom header names & prefix in OAuth2 mode.
• Z-index setting removed from manager/sites to avoid styling conflicts.
• Conversation selector is hidden by default for a cleaner header; you can always enable it in the branding settings.
• The message module allows you to tag AI-generated messages.
• Home View and Roster groups can be linked directly to articles.
• The CleverReach connection has been expanded: forms can now also transfer inactive subscribers directly to CleverReach, for example as part of a double opt-in process. This makes it much easier to manage and automate your email lists.
• The GPT gateway has also been improved. If a request is not answered due to an Azure content filter, it now returns specific information on Azure content filtering. This allows you to understand why a response was blocked and make targeted adjustments.

🆕 Neue Templates im Store

• Event Feedback
• GPT Briefing Chat
• Survey mit Coupon Code
• Your Career Guide
• Contact Search

---

🙏 Thanks for reading!

Your feedback drives every release. Try out the new features, share your ideas and look forward to the next highlights in generative AI & conversational agents. Have fun building!

🌟 April release at LoyJoy: accessibility, AI test automation, intelligent product finder, smart analytics - and much more

🧑‍🦯 LoyJoy complies with the German Accessibility Improvement Act (BFSG)

We are delighted to announce a very special highlight: The LoyJoy chat is now accessible! 🎉 Thanks to comprehensive accessibility optimisations, our chat platform is now fully usable via screen reader, navigable with the keyboard and clearly structured. This means we fulfil the requirements of the Accessibility Reinforcement Act (BFSG), which will apply to digital services from June 2025 - and send a strong signal for digital inclusion and participation. More information in our blog post.

But that’s not all - here are the latest highlights:

📦 New chat widget with web components and ES6 module

Our LoyJoy widget is now provided as a native web component () - integrated via a modern ES6 module. This eliminates the previous iFrame, which enables faster loading times, simpler styling, full CSP compatibility (incl. nonce support) and clean integration into modern web stacks. Ideal for developers and marketing teams alike. 💬🚀 All details in the blog post.

📊 Intelligent Insights (Beta)

Our new Intelligent Insights feature is now available in the Analytics of every Agent. The AI automatically analyses chat performance, creates a report and provides specific optimisation suggestions - ideal for identifying weak points and improving experiences based on data.

🛍️ Intelligent product advisor with GPT

With the new GPT Recommender module, you can build a smart product advisor in just a few minutes - without any programming. Customers describe what they are looking for, the bot asks questions if necessary and suggests suitable products from your product database - including availability, coupon function and add-to-cart (as with Wine in Black, for example).

Product data highlight:

The product database can now also be easily uploaded via CSV or XLSX - for example from your ERP or shop system. This allows you to update product data in just a few seconds - even automatically and regularly.

🧠 GPT Knowledge is even smarter

• PDFs are now automatically divided into structured Markdown sections during upload - ideal for more precise answers.
• When answering questions, the generative AI can now also include the current website on which the user is currently located. In particular, questions that include the context of the current website are answered reliably and without further enquiry.
• GPT is able to rewrite users’ follow-up questions in order to improve the quality of the answers.
• Complex questions can be broken down into sub-questions - this improves the RAG context search and significantly increases the hit probability.

🎨 Further improvements at a glance:

• Display product galleries directly in the live chat
• Delete all products at once in the manager - ideal for large updates
• Emojis in Quick Reply buttons possible 🌟
• Text colour of the header in the chat window as well as the primary and secondary buttons can now be individually adapted

New LoyJoy Release January 2025

We have once again developed the LoyJoy platform to support you in automating, personalizing, and enhancing the efficiency of your interactions.

AI Shopping Agent: Intelligent Product Consultation via Chat

AI-supported product consultation for increased sales will be the big topic in 2025 at LoyJoy!

Our client Wine in Black uses the “Intelligent Sommelier” to help customers discover excellent wines and current offers. Try the Sommelier now

• AI Product Advisor Template: Use our new Product Advisor Template to efficiently implement AI-supported product recommendations.
• Product Manager: Products can now be added and edited directly in the LoyJoy Manager. Learn more
• Products API: Programmatic uploading of product data and integration into processes. Learn more

Gen AI-Supported Knowledge Processing and Explainable AI

• Improved Website Crawler: The HTML of websites is now converted to Markdown to retain the formatting and links of the website. The chat now provides better link recommendations in the responses.
• More Chunks and Optimized Knowledge Integration: In the GPT Knowledge module, configuring the number of knowledge chunks (max. 32) is now possible to receive more relevant answers. If multiple mentions are found in a source, they are now optimally sorted. If re-ranking is enabled in the GPT Knowledge process module, 1/3 of the sources will now always be selected, and 2/3 will be selected based on the re-ranking. This makes source selection more robust.
• Explainable AI – Cite sources directly in the text: With generated answers, sources can now also be cited directly in the text (and not just below the answer). Initially, this will only be supported by the Command-R model from Cohere. To test this, select this model under Knowledge > Tab “Settings”. More models will follow.
• Choose LLM per Module: You can now choose the GPT model in each GPT module separately, allowing for example GPT-4o or o1 selection for a specific module where strong AI is required. GPT-4o-Mini then remains the standard for all other modules.
• Extend Answers with Web Search: The “GPT Knowledge” process module now has the ability to search the internet for sources. The search allows you to search the web for answers to a question in addition to the internal LoyJoy knowledge database. The web search offers different search engines. These can be configured in the module settings.

Extensions in the Chat UI

• Dynamic Text Length in the Chat Footer: The chat footer automatically expands for longer questions and scrolls if necessary. The maximum length of questions can be configured under Branding in the footer.
• “Show All Sources” Function: The chat by default shows the top 4 sources in AI-generated answers but now offers the option to display all sources (simply click the ”+” to expand).
• New API Parameters: With message and suggestions, messages and suggestions for the chat can be pre-defined. For example, the chat can be started from a search input field, and the question entered in the search field is passed to the chat for answering. Learn more

New Process Modules and Functionalities

• Form Module: Create customizable forms with AI-supported validation and autocompletion. Learn more
• GPT Knowledge Improvements: Combine complex API mappings with JSON pointers, XPath, and the LoyJoy Expressions Engine for powerful integration with legacy systems. Learn more
• Brevo Module: We now support direct integration with Brevo! Use personalized email marketing automation.
• New Features for Chat Management: The live chat history can be sent to CRM or ticket systems with the chatHistoryLive function. Learn more

Optimizations for Usability and Design

• Process Editor: Drag scroll with the middle mouse button for easier navigation.
• Chat Design: Colors and input fields in the chat adapt to the brand color.
• Inline Video Playback: Videos in the chat are played inline on mobile devices in the Video Gallery module without switching to fullscreen mode. Learn more

Analytics and Cost Control

• Extended Analytics for Modules: Track API calls for Google Maps in the Places module to better control costs. Learn more
• GPT Cost Tracking: Transparent tracking of AI usage.

You can find a complete list of all changes, updated daily, in our Change Log.

Always a step ahead with LoyJoy – discover the latest opportunities for your business.

Azure GPT-4o Mini, Double the Number of Sources

New Standard Model: Azure GPT-4o Mini

• Integration of Azure GPT-4o Mini as the new standard model for all LoyJoy GPT chats
• Faster, more accurate and ¼ the cost of GPT-3.5 Turbo
• Still using GPT-3.5 for your LoyJoy chats? Then quickly switch to Azure GPT-4o Mini under “Knowledge” → “Settings” → “GPT Model”

Double the Sources With Azure GPT-4o Mini

• The number of sources used by the “GPT Knowledge” process module from your knowledge base is doubled
• This allows for even more accurate answers than before
• It is enabled for newer LLMs like Azure GPT-4o Mini

New Name Module, IP Address Control, Combobox and Support for Kazakh

New Process Module: Name

• Combining first name and last name in a single module
• Collect first and last names in a single step

IP Address Management

• Gain more control over your bot’s accessibility with our new IP address feature. You can now add specific IP addresses or IP address ranges for access or exclude certain IPs to improve security and targeting options

Combobox in Questionnaires

• We’ve upgraded the Questionnaire module with a searchable dropdown (Combobox) option. This feature makes it easier for users to find and select options in long lists

New Language: Kazakh

• We’ve added Kazakh as our 39th supported language

Three New LLMs in One Day: Llama 3.1 405b, GPT-4o Mini and Azure GPT-4o, Hosted in the EU, Now Available in LoyJoy

Llama 3.1 405b:

• 405 billion parameters for in-depth understanding
• Strong performance in coding and text analysis
• More cost-effective alternative to GPT-4o

GPT-4o Mini:

• Outperforms GPT-3.5 in important benchmarks
• 60% cheaper and twice as fast as GPT-3.5 Turbo
• Ideal for fast, cost-sensitive applications
• Compact size with high performance

Azure GPT-4o:

• The most powerful LLM GPT-4o to date with increased data security
• Full EU hosting now available for maximum GDPR compliance
• The Sweden location is available

Llama-3 70b: Generative AI 100% made in Germany

The new integration of Meta’s open-weight model Llama-3 70b into the LoyJoy Conversational Platform offers a new LLM which is hosted by our German partner primeLine AI in Limburg an der Lahn, Germany and ensures that everything is 100% made in Germany.

Interested? Contact us for activation.

Key Features

• Hosted in an ISO 27001 certified data center of partimus GmbH in Limburg an der Lahn (part of the primeLine Group)
• Privacy focused: Data stays in Germany
• No Azure OpenAI hosting required
• 100% Made in Germany
• As always, full GDPR compliance

New LoyJoy Feature: Reranking

Our new Reranking feature is an optional RAG enhancement and improves chatbot responses by intelligently analyzing and prioritizing up to 32 information sources per request. The result is more accurate, relevant, and comprehensive answers.

New LLM: Claude 3.5 Sonnet

We’ve integrated Claude 3.5 Sonnet, Anthropic’s latest AI model, offering superior performance in reasoning, code analysis, and multilingual math, along with improved visual capabilities and cost efficiency.

New LLM Integration: Claude-3 & Llama-3

We have added the two LLMs Claude-3 and Llama-3 to the LoyJoy Platform. You can now use Claude-3 and Llama-3 in your experiences.

OAuth: Super Easy Google Sheets Connection

Google Sheets can now be connected more easily. In the Google Maps and Product Feed modules, Google Sheets can now be authorised as a connector even more easily. Simply connect with your Google account and insert the spreadsheet ID. No key is required for authorisation.

SQL Parser

You can now use SQL to filter data from connectors. Specify the spreadsheet ID and now you can find data even more easily from large amounts of data.

New: GPT-Source Folder

You can now organise your GPT sources in folders. This makes it easier to select the right sources for different experiences.

Tag Suggestions & Questionnaire

This release introduces a new feature that allows you to add tags to answers in the Questionnaire module. This will help you categorize and filter answers more easily and improve your product recommender.

Salesforce Apex API Client

You can now easily integrate your Salesforce Service Cloud account with the Salesforce Service Cloud module. With the new mode “API request”, you can retrieve and/or send data from your Salesforce account on the fly. In addition, you can now use Oauth authorization to easily connect to your Salesforce account.

JSON Validation

We have also released a new JSON validation feature that allows you to validate your JSON code. You can find it in the API client module and in the Salesforce Service Cloud module under Payload.

New Sources API

This release adds a new API that allows you to add sources to be used by GPT. The API can be used to add websites as sources or to upload files such as PDFs. You can use the API to automatically remove or add sources. This allows you to keep your sources up to date without manual intervention.

New Articles Module, Backend Improvements

The new articles module allows you to display selected articles directly in the chat flow. This allows you to highlight certain articles e.g. as an FAQ or to provide an overview over the topics you cover.

Linking directly to articles

Additionally, you can now link directly to articles from the chat flow. This is implemented in a new link selector that can be used in any text field.

LoyJoy Backend Improvements

• When uploading images, you can now zoom out to e.g. fit a portrait image into a landscape aspect ratio
• You can switch between sorting by name and sorting by date in the list of experiences

LoyJoyGPT 2.0

This release completely overhauls the GPT integration in LoyJoy. There are new GPT modules such as the GPT follow-up question, GPT Prompt, GPT Smalltalk, and GPT Gateway. The previous GPT integration can be found in the GPT Knowledge module. The new GPT modules are more powerful and flexible than the previous integration and allow you to create more complex and dynamic conversations.