OMQ Update | Version Montevideo 10.5 - Voice Bot, New Chatbot Widget and Answer Pipeline

We have released product version Montevideo 10.5.

With this update we are introducing three major innovations: the OMQ Voice Bot, with which we now also cover the phone channel, our new chatbot widget with Proactive Mode for web and app, and a new answer pipeline with the option to connect your own systems via MCP. This lets you deliver even better customer service across all channels and reach your customers exactly where they need help.

In this article we show you how these new highlights work.

OMQ Voice Bot: Your customer service is now automated on the phone too

With the OMQ Voice Bot you can now automate your customer service on the phone as well. It draws on the same knowledge base as your other OMQ products and integrates seamlessly into your existing phone system: one knowledge, all channels.

The highlights at a glance:

• Communicates and answers in natural language, usable in multiple languages
• Recognises intent even with vague or imprecise phrasing
• Answers are based on your existing service knowledge
• Deployment scenarios and service hours are individually configurable
• Recognises the matter and, where needed, transfers the call to the right contact person or extension
• Messages and callback requests can be captured and forwarded in a structured way

An individual, step-by-step rollout plan is possible and will be accompanied by us.

New chatbot widget: Proactive Mode addresses visitors directly

With this update we are releasing our new, in-house chatbot widget. One highlight: Proactive Mode. Instead of just waiting for a click, the chatbot now addresses visitors on its own, with a short, rotating message that you can configure individually for your chatbot.

One click opens the full chat widget, where customers can select a topic or ask the chatbot a question directly. The open chat widget can also be individually configured. Otherwise, the chatbot answers questions as usual.

And the chatbot widget can now be embedded not only on your website but also directly in your own iOS or Android app. The chatbot in the app draws on the same service knowledge as on the website. Customers simply stay in the app and get the help they need right there.

Enhanced answer pipeline: even more precise, more flexible and now with MCP connectivity

With this update we have enhanced our answer pipeline, making it even more precise and flexible: depending on what is needed, it searches multiple times to gather all the information required for an answer, even when it is spread across several documents or pages. If no sufficient answer exists, that is communicated openly.

What is MCP?

MCP (Model Context Protocol) is an open standard that enables an AI not just to talk about other software, but to work with it directly. Previously, every connection between an AI system and a business application had to be built individually. MCP replaces that with a common language: a system publishes what it can do, for example “retrieve order status”, “check appointment” or “create ticket”, and any AI that speaks MCP can use it, without a custom-built integration.

MCP at OMQ: connecting your own systems

And that is now possible with our OMQ system: with the help of our OMQ AI Agents, the chatbot can decide during the conversation which system it needs to access, retrieve the right information and deliver the answer. An MCP connection is simply another access route, for example the actual ticket status from your helpdesk, the order status from your shop, a free slot in the calendar, a contract detail or an address change in the CRM, plus many further integration options.

Further new changes

OMQ Help

• Accessibility was improved in several further steps, including the items from an external accessibility audit, the print button and empty FAQ headings that previously produced an empty heading element.
• Search engine optimisation was improved in several places: deleted questions no longer return a success status, the canonical URL no longer contains the category, file paths return a noindex header, FAQ snippets are now also returned on single-question pages including breadcrumb structure, and the output of structured data can be disabled per integration.
• It is now possible to set the page title.
• A category can now be preselected in the Help client.
• Entries can now be excluded from the list of most frequent questions, and questions can be pinned there. The exclusion now applies to all entries of a category, optionally including subcategories.
• OMQ Help can now also be embedded in sealed-off environments such as Confluence, via a new setting that suppresses writing the URL and the link targets.
• Server-side rendering and the geo functions were improved, several errors were fixed.
• Several improvements on mobile devices: the icon in the search field remains a magnifying glass, the search no longer closes when Enter is pressed without a selection, and the close icon is visible again on desktop.
• A loading screen is now shown when entries are reloaded, subcategories can be shown via a setting, and the category icons are larger in the default style.
• A problem was fixed where answers were shown in the wrong category. Also fixed: errors when no category was selected, when a linked question was deleted and when a non-existent question URL was opened. The latter now redirects to the main path.
• Source code elements inside conditions were displayed collapsed. This problem was fixed, as was an intermittently invisible print icon.
• The search is more robust against unusually complex queries and deeply nested conditions that could previously lead to timeouts.

OMQ Chatbot

• A completely new chatbot client was implemented: with its own mobile interface, isolation from the styles of the surrounding page, styling loaded at startup and support for use in iOS and Android apps.
• The accessibility of the chatbot client was reviewed and improved.
• Numerous improvements to look and behaviour were implemented over several rounds: opening and closing animations, a typewriter animation for welcome texts, automatic detection of links in text, positioning (left, centre, right), logo upload from the organisation, file upload with progress indicator, an interaction reminder and many smaller layout and styling fixes.
• Delays before the first message were reduced, and a session is restored more reliably after closing and reopening.
• The session is now stored per connection instead of per domain. Multiple chatbots on different pages of the same domain therefore no longer restore each other’s conversations.
• Used sources can now be displayed in the chatbot; suppressed or unused sources are no longer listed.
• Proactive Mode was reworked.
• An AGENT integration was added, through which chats can be conducted by AI agents.
• If a requested language is not enabled, the chatbot now falls back to a configurable primary language.
• Fixed: problems with navigation via menu entries, action buttons without a function, circular redirects and conversation updates without an external ID.
• Error messages of the Userlike integration are now more meaningful, and the icon was updated.
• Menu search requests can no longer overload the server.

OMQ Voice

• OMQ Voice is new: a voice bot for phone and web that answers with the same knowledge foundation as the chatbot and the other products.
• Voice bots are their own unit with their own administration interface, including voice selection, configuration of the speech output and a configurable transfer destination.
• The phone connection runs via sipgate, including sipgate AI Flow.
• Input via the phone keypad is supported, including multi-digit input with a timeout per keypress.
• Knowledge sources are available to the voice bot, so uploaded files and synchronised websites can be used in the conversation.
• Links from answers are passed to the voice bot so they can be referenced during the call.
• A voice web widget is available, with its own public key and a configurable request limit per voice bot.
• Voice conversations appear in tracking and reporting, including the time spent in the conversation.

OMQ Contact

• Accessibility and presentation benefit from the reworked style system that OMQ Contact shares with OMQ Help.
• A problem was fixed where contact information with non-text elements led to a processing error.

OMQ Assist

• A problem was fixed where OMQ Assist reported an error and did not start.
• OMQ Assist works again with the current Znuny version, which had previously made the plugin unusable.

OMQ Reply

• A new pipeline was implemented for OMQ Reply that works with tools and can use automators and connected systems directly while writing an answer.
• Email answers are now generated via a configurable email template, and the templates are managed centrally.
• The email pipeline now takes the entire email history into account and additionally uses the subject of the message.
• Knowledge sources can now be used in OMQ Reply, and you can choose whether the knowledge base or the knowledge sources take priority.
• Multilingual support was implemented: a connection can now serve several languages, and the answer opens in the language of the enquiry.
• Custom triggers were implemented for OMQ Reply, including an interface, pre- and post-processing steps and the option to append additional output.
• Conversations can now be viewed in the administration, and sent messages appear in the conversation instead of an internal status.
• New and extended integrations: OMQ Reply for Zendesk with its own interface, Userlike, Greyhound, multiple answers for OTRS, plus numerous fixes for OTRS, Znuny and Zammad, including output as plain text instead of HTML, correct answer templates and no answer when an agent is already assigned.
• Email connections can now be tested, copied and sent via a separate sender address; multiple addresses can be configured for Exchange connections.
• Expressions of thanks are now recognised and treated as resolved instead of triggering an answer.
• Numerous problems in email processing were fixed: messages in unusual character encodings, nested multipart messages, embedded images, multiple recipients, empty subject lines, missing text content, collapsed or duplicated line breaks in quotes and oversized requests to connected systems.
• A single faulty connection no longer prevents the processing of the remaining connections of an account, and skipped messages were corrected.
• A problem was fixed where a draft marked the original email as answered even though the draft was discarded.

OMQ Automator

• Automators can now be used as tools of the answer pipeline and no longer need to be stored separately for this.
• An automator for Python code was added.
• A mode was added in which a form is sent without a confirmation step.
• A multi-select element is now available, and forms now also transfer the values of surrounding conditions.
• A form input can now be exited again.
• Automators in the chatbot were improved; a form without a defined submit button is now sent automatically.
• Email automators are now also sent when the reply-to field is empty.
• A problem was fixed where editing an automator from the knowledge base displayed an error.

OMQ Engine

• A new answer pipeline was implemented in several stages. The system now decides for itself which tools it needs to answer a question, search in the knowledge base, search in knowledge sources, automators or queries to connected systems, instead of following a fixed sequence.
• OMQ now operates the file search for knowledge sources itself instead of using the language model provider’s file search. That gives better control over indexing and results and combines keyword and semantic search in one result list.
• The connection to the language models was migrated to the current interface, and reasoning models are now supported.
• The answer pipeline can now determine the current date and time itself when a question requires it.
• The priority between knowledge base and knowledge sources is now applied when composing the answer, and both sources are always consulted.
• Answers that are not covered by the existing knowledge are now returned as empty instead of freely phrased, and the number of source references is limited.
• The handling of links in answers was significantly improved: damaged, truncated and incorrectly rewritten links are detected and corrected, and pages from synchronised websites are linked directly.
• Documents are processed more reliably: tables from Excel and CSV files are converted into a form readable by the language model, PDF files are supported in website synchronisation, and websites that are only built in the browser can now be read.
• Website synchronisation was fundamentally reworked: a queue-based daily update, crawl delays from robots.txt, exclusion of individual URLs, support for manually stored sitemaps, and correct handling of redirects, certificates and URLs with parameters.
• Indexing has become significantly faster and no longer allows a single account to block all others. Restrictions for very large knowledge bases were lifted.
• Language and text processing were improved: word separation, list numbering, Markdown processing, code blocks and the preservation of spaces around links during translation.
• Content blocked by the language model provider is now handled cleanly instead of triggering an error, and temporary overloads are retried automatically.
• Knowledge sources in draft status are no longer used for answering, and duplicates in source lists were prevented.
• The OMQ knowledge base can now be provided as a service for AI agents, and external tools can be connected, with authentication, so the answer pipeline can use them.

Administration

• The style system was fundamentally reworked. In addition, invalid CSS no longer prevents saving, and the style preview is displayed immediately.
• The administration menu is now adjustable in width.
• The knowledge base was extended: an interface for the export, faster loading of categories, deletion of categories including sub-elements, an improved category selection and search results whose link can be shared.
• An HTML snippet element was added to the document editor. YouTube Shorts now work, and YouTube videos can be embedded without cookies.
• Knowledge sources: the file of a file knowledge source can be replaced and downloaded, individual URLs can be excluded, a warning is shown for empty or unreachable sitemaps, and loading for accounts with many knowledge sources has become significantly faster.
• Reporting was significantly extended: knowledge sources appear next to the questions and in the category view, OMQ Reply conversations can be viewed, the email subject is included in the content, the export contains the number of responses, and monthly reports are available. Reports are now also accessible to editors.
• The administration of OMQ Reply received a search function, and connections can be copied and tested.
• Custom instructions and custom triggers were opened for selected accounts and have their own interface for the chatbot and OMQ Reply. When creating a new connection, a sensible default instruction is now preset, and the maximum length was significantly increased.
• An interface for managing tools and connected systems was added.
• The interface now indicates when the AI is working on an answer and when indexing is running.
• Usage and report data can now be provided via a read-only interface with modern authentication for AI agents.
• Numerous smaller fixes: saving no longer jumps to the first tab, text selection works again in Chrome-based browsers, image upload works again, an error is shown when the file size limit is exceeded, an answer can have a start date without an expiry date, and the dialog for reporting problems no longer loses entered content.

General

• Passwords are now stored with a modern hashing method. Existing passwords remain valid, a reset is not required.
• After several failed login attempts, an account is temporarily locked.
• Two-factor authentication can now be activated for individual users instead of the entire account. Several problems with two-factor authentication were fixed.
• A problem with the password reset email was fixed, including for addresses with different capitalisation.
• Automated access to knowledge base entries without a browser is now detected and blocked.
• The caching of frontend files was corrected so that no outdated files are used after a release.
• The search infrastructure was updated to a newer version, as were the frontend libraries.
• Stability and operations were improved in many small steps: longer time limits for long-running requests, significantly less noise in the logs, better monitoring of memory consumption and fixes when deleting accounts and users.