Skip to main content

RETORIK Changelog

Release Note v1.21.0-preview​

Available on the Preview channel only.

New LLM-based intent recognition engine​

The user intent recognition engine has been completely rebuilt. It now relies on an LLM (language model), which significantly improves the understanding of user requests. Example phrases are no longer required in intents: simply fill in a description field to explain the role of each intent to the model.

Redesigned administration interface​

This version marks the beginning of a global reorganisation of Retorik Studio for a simpler, more accessible experience. The following sections have been reviewed and enhanced:

Digital employee — brings together identity configuration (character, name, voice, language), knowledge (role, generative AI, conversational style) and the welcome screen (welcome message, default mode).

Data sources — allows indexing of documents, files and content items to feed the agent's responses. Each source can be enabled, disabled, filtered by tag or associated with a type of semantic search.

Knowledge — contains the knowledge base (FAQ) and dictionaries. This is where questions/answers and training phrases for the intent recognition engine are defined.

User interface — brings together design (logo, background, colours), navigation menu configuration and integration type (full screen, widget, mobile).

Relaunches — configures messages sent automatically after a period of user inactivity.

Deployment — groups the technical integration parameters (scripts, CSS, URLs) for embedding the agent into a website or third-party application.

Comments — provides access to feedback left by users at the end of conversations, useful for evaluating response quality.

News — management of the news ticker content displayed continuously by the agent.

Statistics & History — usage dashboard, detailed conversation history and agent performance analysis.

GDPR compliance — tools for managing consent, data anonymisation and deletion requests.

Other new features​

Re-indexation warning — when editing a data source, a visual warning now indicates whether re-indexation is required for the changes to be taken into account.

Form colours — the interface customisation page now features a dedicated section for configuring form colours, with an integrated colour picker.

Validation of questions before satisfaction — it is no longer possible to enter a negative value for this parameter. An explicit error message is displayed if attempted.

Improvements and Bug Fixes​

  • Optimisation of chunk assembly in streaming mode (reduced display latency).
  • Improved handling of attachments in streaming within the ChatGPT dialog.
  • Token estimation for LLM providers that do not return usage information.

Release Note May 26, 2026​

Retorik Studio: Version 1.20.18

Improvements and Bug Fixes​

  • Improvement of documentary source management:
    • During a conversation with the bot, it relies on documents to build its responses. Some of these documents are displayed to the user as sources, while others are used behind the scenes without being shown.
    • Before: the bot could sometimes fail to display sources even though it had used some, because the distinction between "used source" and "displayable source" was not correctly handled.
    • Now: the bot clearly distinguishes between the two. The sources displayed to the user remain the same as before — only the reliability of the internal mechanism has been improved.
    • No action required on your part. This update is transparent.

Release Note May 22, 2026​

Retorik Studio: Version 1.20.17

New Features​

  • Sources displayed directly in the bot's response:
    • Previously, documentary sources appeared only at the end of the response, as a separate block. From now on, references to sources are inserted directly in the response text, throughout the conversation, in a more natural and contextual way — like integrated footnotes.

Improvements and Bug Fixes​

  • Improved platform monitoring: internal improvements have been made to better monitor and optimise platform performance.

Release Note April 30, 2026​

Retorik Studio: Version 1.20.16

Improvements and Bug Fixes​

  • Fix for localisation in suggestion cards:
    • A text displayed in suggestion cards was not correctly translated according to the user's language. The display now respects the language configured for each user.

Release Note April 23, 2026​

Retorik Studio: Version 1.20.15

Improvements and Bug Fixes​

  • Improved deployment reliability: the platform installation process could sometimes fail due to synchronisation delays. Verification and automatic retry mechanisms have been put in place to make this process more robust.

Release Note April 2, 2026​

Retorik Studio: Version 1.20.14

Improvements and Bug Fixes​

  • Fix for configuration settings save (Setup): saving settings in the configuration wizard could silently fail under certain conditions. Improvements have been made to make this mechanism more reliable and ensure that settings are properly saved.

Release Note February 24, 2026​

Retorik Studio: Version 1.20.13

Improvements and Bug Fixes​

  • Update of platform technical components: the underlying technical foundations have been updated to their latest stable versions. These updates bring improvements in performance, security and reliability, with no visible impact on end users.
  • Platform cleanup and stabilisation: inactive components have been removed and deployment anomalies have been fixed, contributing to a lighter and more stable platform.
note

Version 1.20.12 is the first version available in the git history — no comparative notes available.

Release Note November 3, 2025​

Retorik Studio: Version 1.20.8
Retorik Framework: Version 3.2.5

New Features​

  • New interface
    • The new interface is now enabled by default for new projects (Retorik Studio tenants).
    • Existing deployed clients can choose whether to switch.
    • The activation banner for the new interface has been removed from preview.mybot.retorik.ai and mybot.retorik.ai.
  • Transport skill
    This skill, dedicated to public transport, allows:
    • Querying the digital employee in natural language about public transport data from standardised GTFS (General Transit Feed Specification) static and real-time formats. Examples of available information: schedules, stops, routes, delays & service alerts, fares, accessibility, etc.
    • Dynamically displaying alerts in a digital employee's news ticker.
  • Audit log
    • Connections to the agent and settings made in Retorik Studio by authorised users are now chronologically recorded in an audit log, available in the Audit Trail menu.
    • Logs are retained for 10 days by default.
    • Settings are reserved for DAVI administrators. See "Configuration > Settings > Audit Trail".

Improvements and Bug Fixes​

Improvements to Retorik Studio:

  • Integration
    • Ability to add script and CSS file URLs to customise the agent's interface when integrating it into a website or other applications (e.g. La Vitre).
  • News ticker:
    • Customisation of the red "EN CONTINU" banner text.
    • Option to not read the title and/or subtitle of the news ticker banner.

Bug fixes in Retorik Framework:

  • Home:
    • Fix for size and missing image when only one tile is defined on the home screen.
    • Incorrect URL assigned to the Retorik logo: changed to https://davi.ai / Click disabled on Kiosk.
    • Mobile/Widget: the application does not start if the user does not switch to voice mode at least once.
  • Speech recognition:
    • Voice transcription was truncated from the 3rd line on Desktop. From now on, and only on Desktop, when the user's dialogue is long, 2 or 3 lines disappear and the rest of the dialogue starts with 3 dots followed by the last word of the previous sentence, to avoid losing track of the speech in progress.
    • Unwanted display of the "Press the microphone" bubble even though the option "Activate microphone when the agent is waiting for a response" is set to "Always".
  • Speech synthesis:
    • After 10 minutes, responses are no longer read by the agent but logged in the history.
    • Issue with sentences being read twice.
    • The end of the read sentence is not consistent with the text response.
    • The option "Enable subtitles in voice mode" does not work.
    • Subtitle positioning now customisable on an 86-inch vertical screen (e.g. La Vitre).
  • News ticker:
    • Banner positioning now customisable on an 86-inch vertical screen (e.g. La Vitre).
    • Removal of the zero displayed in the news image area when no media is defined.
  • Relaunches:
    • Welcome relaunch became random when the GDPR module was activated.
    • Relaunches not sent in Kiosk mode.
  • RTL mode:
    • In Arabic, the character must remain on the left and the remote control on the right, as in other languages. Only the tiles and their text should be displayed right to left.
  • Anghello:
    • First name and last name not aligned when the company is not detected.

Release Note September 22, 2025​

Retorik Studio: Version 1.20.7
Retorik Framework: Version 3.0.1

New Features​

  • Custom menu
    It is now possible to customise the menu by selecting features, sorting them as you wish, by category or not, and customising the title, pictogram and colours.
    Discover this feature in the "Digital Employee > Settings > Menu" section of Retorik Studio. It works similarly to the click-to-action creation available in the home settings.

Release Note July 16, 2025​

Retorik Studio: Version 1.20.5
Retorik Framework: Version 2.1.0

New features and bug fixes moved from the Preview channel to the Production channel.
Test banner for new features available only via the "Test" button in Retorik Studio (url mybot.retorik.ai/project name)

Bug Fixes​

  • History > Dissatisfaction comments: the original question has been restored.

Release Note July 7, 2025​

Available on the Preview channel only.
Retorik Studio: Version 1.21.0
Retorik Framework: Version 2.1.0

New Features​

  • Option to switch to the new interface
    Discover our new remote control and a more intuitive voice mode in preview:

    • Temporarily, by clicking the "Try it now" button on the welcome banner. The test only applies to the current conversation. The banner appears for each new conversation with the digital employee.
    • Permanently, by enabling a setting in the Retorik Studio backoffice ("Digital Employee > Settings > Integration > General Settings", option "Enable new interface").

  • Digital employee with Generative AI:

    • Ability to increase the number of documents from the semantic search applied to data sources (limited to 5 in the previous version) and to increase the prompt size received by the AI (between 6,000 and 100,000 tokens vs. 6,000 tokens previously).
    • Ability to create a synonyms dictionary to improve semantic search.
    • Implementation of satisfaction measurement for AI responses. For configuration, refer to the Satisfaction section.
    • Ability to set up a list of questions suggested by the AI, related to the current discussion. Configure a relaunch in "Digital Employee > Relaunches" with an empty trigger (applicable to all messages) and the following instruction in the message field (to be customised):
      [generatedsuggestions text="I can suggest these questions related to our discussion" min=2 max=4]

  • Voice mode > Microphone management: ability to activate the microphone when the agent is waiting for a response, with 3 modes:

    • Automatic: the user must click the microphone to speak, except for functions where Retorik provides automatic microphone opening, such as satisfaction measurement, the Anghello feature entry and exit process, and form-collected data defined with automatic opening.
    • Always: the microphone opens automatically after each agent message.
    • Never: the user must always click the microphone to speak.

Bug Fixes​

  • Conversation history > CSV export: fix for anomalies encountered when selecting time slots.

Release Note June 23, 2025​

Retorik Studio: Version 1.20.1
Retorik Framework: Version 2.0.4

New Features​

  • Quick start for a digital employee:
    After creating a new project on https://retorik.portal.ai and first login to Retorik AI by an account defined as owner in the organisation, the user is taken through a quick setup of their digital employee in 6 steps:
    • Identity: character, first name, role, voice, language and function
    • Knowledge: configure generative AI with an AI service (e.g. Retorik conversation, Azure OpenAI or OpenAI), a conversation style, knowledge domains and forbidden topics.
    • Welcome: default conversation mode (voice or text) and the welcome message delivered by the digital employee.
    • Interface and brand guidelines: logo, colours, background.
    • Integration type: full screen or widget mode (desktop or mobile).
    • Digital employee preview: validate the user interface. Option to return to any configuration step to change choices, until the preview is validated. This configuration ends with the "Access advanced settings" button: the user accesses the backoffice menu adapted to their role.

Bug Fixes​

  • Conversation history > CSV export: fix for formatting issue in the "Comments" column when it contained an AI prompt, and limit the prompt to the character limit accepted by an Excel cell (32,767).
  • User interface — fixes for:
    • icon size in the discovery card,
    • image display in home tiles,
    • user text colour in history,
    • window depth issues with overlapping windows,
    • incorrect display of the close button in kiosk mode.

Release Note April 24, 2025​

Retorik Framework: Version 2.0.0

Improved User Experience​

  • Redesign of text mode:

    • A smoother and more natural display for more pleasant and efficient exchanges.
    • A more structured speech from the digital employee.

  • RTL mode (Right To Left) support for languages:

    • 'ar': Arabic
    • 'fa': Farsi
    • 'he': Hebrew
    • 'ps': Pashto
    • 'ur': Urdu

New Features​

  • Integration of the Anghello feature dedicated to visitor management:
    • Easily record the arrivals and departures of your visitors in your company.
    • Raise awareness among your visitors about your safety instructions.
    • Notify your team in real time.
    • Details of this feature will be available in the Retorik documentation at the end of July / beginning of August 2025. For more information, please contact your DAVI representative before that date.

Release Note February 3, 2025​

Retorik Studio: Version 1.18.0

Technical Updates​

  • Implementation of single sign-on for all Retorik services.
  • Implementation of the Retorik Portal service, allowing authorised Retorik accounts to manage their digital employees and autonomously create a new employee (service reserved for DAVI staff in this version).

Bug Fixes / Improvements​

  • Dashboard > Summary: not all "UserMessage" type messages were included in the "User messages" column. Examples: messages processed by AI.
  • Localisation > Culture export: fix for the URL to access the ZIP file generated when exporting language texts. Copy-pasting this URL into a browser is no longer necessary: a here link has been added.
  • Home > Click-to-action > Pictograms:
    • Fix for colour issues in the Retorik Studio preview and user interface.
    • Option to not apply the colour filter in order to keep the original colour of a pictogram.
    • Removal of the "Hospitality" and "Media" groupings from the dropdown list.
  • Data sources > Edit: option to correct the "Filter" field.

Release Note November 25, 2024​

Retorik Map: Version 1.1.0
Note: this package must be integrated into a web application using version v1.1.7 of the Retorik Framework.

New Features / Improvements​

Interactive map:

  • Added the distance banner by transport mode and the [Go] button at the bottom of a geomarker card. The route is also drawn on the interactive map (similar to the detail view).
  • Optimised geomarker display time.
  • Increased zoom level by 4 points to better distinguish very close geomarkers that may overlap.

Technical Updates​

  • When adding Retorik Map to the Retorik Framework menu, it is now possible to import the "labels" object from the Retorik Map package, which contains translations of "Interactive map" in all languages supported by the Retorik Framework.

An example menu is shown below (taken from the "example" directory of the Retorik.Map git repository):

Code

import { MapIcon, Map, labels } from '@davi-ai/retorik-map'

baseMenu: [
      { view: AvailableViews.home, indice: 1 },
      {
        customDisplay: 'mapDisplay',
        label: labels,
        icon: <MapIcon />,
        indice: 2
      },
      { view: AvailableViews.news, indice: 3 },
      { view: AvailableViews.weather, indice: 4 },
      { view: AvailableViews.emergency, indice: 5 }
    ],

  • The Retorik Map package is now available on npmjs, in addition to the Retorik feed.

Release Note November 6, 2024​

Retorik Studio: Version 1.17.2
Retorik Framework: Version 1.1.7

Bug Fixes / Improvements​

Data sources: memory optimisation during indexing processing.
Relaunches: fix for pictograms not displayed on click-to-action items during a relaunch with home recall in grid mode.
AI responses with streaming enabled:

  • Fix for duplicated responses in certain cases.
  • Fix for duplicate media card display when using image-type URLs.

Release Note October 9, 2024​

Retorik Studio: Version 1.17.0
Retorik Framework: Version 1.1.6
Retorik Kiosk: Version 1.1.4

New Features​

  • General: the "Chatbot" feature becomes "Digital Employee". All terms "chatbot", "agent" and "assistant" are replaced by this new terminology.

  • Home: customise the home screen with click-to-action items:

    • Up to 9 click-to-action items, displayed in a 3×3 grid format. Each item is represented by a customisable tile with a Title, Pictogram and Colours.

    • Each click-to-action item is associated with a type of information to display, selected from a dropdown list. Examples:

      • A question or questions,
      • A list of content items, a content recommendation carousel or a content detail card,
      • A link to a web page,
      • The interactive map, weather, etc.

    • In widget and smartphone mode, tiles are displayed in a carousel.

    • This grid of click-to-action items can be recalled in a relaunch using the shortcode [tilecard] or [tuiles].

    • This feature is not compatible with the use of shortcodes in the welcome message.

  • Knowledge > Data sources: option to filter a Content-type data source to limit semantic search and AI responses to more targeted content.

Bug Fixes / Improvements​

  • Digital Employee > Settings > Integration: fix for the "Enable subtitles in voice mode" parameter that was no longer working.
  • "Hospitality" feature: the list of content filters and their values, created by default when the feature is activated for Askléa (our digital employee dedicated to tourism), have been cleaned up to leave only standard elements. This fix only applies to new tenants.
  • Display of the Retorik Framework version in the bottom left of the MENU page of the digital employee interface.

Release Note September 19, 2024​

Retorik Studio: Version 1.16.11 (22/08/24)
Retorik Framework: Version 1.1.5 (19/09/24)
Retorik Kiosk: Version 1.1.3 (18/09/24)

New Features​

  • A loading animation now appears if the agent's response takes more than 1 second to process.

Technical Updates​

  • Reduced Framework package size.
  • Lucene indexing replaced by ElasticSearch indexing for improved performance:
    • On the Retorik Studio side: faster search in the following features: Knowledge, Dictionaries, General conversation and Conversation history.
    • On the Retorik Framework side, Hospitality feature: faster display and filtering of content items in lists and recommendations on the "Discover" home card.

Bug Fixes / Improvements​

  • Fix for the interactive map not displaying when the digital assistant was queried in conversational mode a second consecutive time.
  • Fix for the POI card display issue on a geomarker in the interactive map (Smartphone).
  • <br/> tags placed in a content item description are now processed in the detail card, the POI card accompanying an agent response, and in the PDF downloaded from the detail card.
  • Optimisations in the processing of the data source indexing queue.
  • Bug fixes identified in Retorik Framework versions v1.1.x prior to v1.1.5: bullet point formatting issue in AI responses, error launching the interactive map, display issue with lists when called successively from the menu.

Release Note July 4, 2024​

Retorik Studio: Version 1.16.0
Retorik Framework: Version 1.0.18
Retorik Kiosk: Version 1.1.2

Technical Updates​

  • Generative AI: in the "History > Conversation" feature, in the technical information accessible by clicking on a response, the LLM configuration is now added alongside the prompt received by the LLM. Additional information logged: Service, Endpoint, Model, Number of messages retained in the AI context, Response length, Temperature and Top P.
  • Semantic Recognizer feature (Experimental):
    • Similar to the "Deep" feature, this feature recognises intents and entities defined throughout the backoffice but using the LLM you have configured. Using an LLM requires fewer training phrases and improves overall agent understanding. Recognition time is also improved, enabling smoother exchanges.
    • Prerequisites:
      • Disable the "Deep" feature,
      • Enable the "Semantic Recognizer" feature,
      • Re-train the agent.

Bug Fixes / Improvements​

  • Generative AI enabled with the streaming function:
    • The "Display starts, then blank, then full message" issue is resolved.
  • Widget format on mobile: sound muted on various user actions: closing the widget, changing page in the browser, minimising the application, putting the phone on standby.
  • "Hospitality" feature, from a home card with custom filters:
    • CATEGORY issue at the top of the POI card.
    • POI display issue in a foreign language on a list activated in French and vice versa.
    • Import of standard POIs from a DecibellesData feed (Restaurants, Accommodation, Activities, Attractions, Products, Services, Events):
      • POIs without "Name, Description, Category, Latitude and Longitude" are no longer imported to improve the quality of data presented to the user.
  • Data sources:
    • View documents: added pagination to maintain display performance for large volumes.
    • Download: if file size > 100 MB (limit for 1 batch of files), the user is notified.
  • Chatbot > Settings > Integration: removal of the language selection, which became obsolete.
  • NLP Deep:
    • Added a processing start notification, in addition to the existing end-of-processing notification.
    • If training remains stuck for 30 minutes, a new training is automatically relaunched. If this second one fails, a training failure notification is sent, identifying the model and skill that failed.
  • Kiosk application (Kiosk mode): configuration changes not taken into account instantly.

Release Note June 19, 2024​

Retorik Framework: Version 1.0.17

New Features​

  • Loading animation:

    • On standard cards:
      • Link-type cards: list of questions for example.
      • Yes/No card: additional response for example.
    • On "Hospitality" feature cards:
      • "Category" button on the "Discover" home card.
      • "Learn more" button on a "Content item" card.
    • On form cards: "Submit" button.

  • Change in Framework parameter application:
    In previous versions, if a manual Framework configuration existed, the settings defined in Retorik Studio ("Chatbot > Integration" menu) were not taken into account — the system was exclusive. From now on, the system takes into account Retorik Studio settings + specific settings according to these new priorities:
    1 — Default parameters defined in the Framework Config file.
    2 — Retorik Studio parameters.
    3 — Parameters passed specifically at Framework level.

Bug Fixes​

  • "Discover" home card: titles that are too long overflow the frame.
  • Kiosk application (Kiosk device): the home screen does not launch when waking from standby.
  • Fix so that the interactive map works with a territory outside the AskLéa application.

Release Note May 28, 2024​

Retorik Studio: Version 1.15.3
Retorik Framework: Version 1.0.16

New Features​

  • Ability to customise the pictograms of the DiscoverCard home card (applicable with content items) by specifying the full URL to access the image (the image can be stored in the media folder or available from another URL).

Bug Fixes​

  • Switching from Widget to full-screen mode and vice versa: fix for duplicate sentences in agent responses.
  • Kiosk application (Kiosk mode): fix on the interactive map for the POI card display issue from the "Learn more" button on the media card displayed when clicking a geomarker.
  • AskLéa web application: POI display issue on the interactive map.

Release Note April 30, 2024​

Retorik Studio: Version 1.15.0
Retorik Framework: Version 1.0.15
Retorik Kiosk: Version 1.1.1

New Features​

"Relaunches" feature:

  • This functionality existed in the previous version of Retorik Studio for 3 trigger events: after the welcome, after the misunderstanding message and after each agent response. This new version offers 10 additional trigger events, such as "after closing a list of content items" or "after closing the conversation history".
    • This allows the welcome card to be redisplayed using the same shortcode defined in the welcome message, after a period of inactivity on the part of the user (e.g. 5 seconds). Please note that the shortcode must be preceded by an introductory text for the relaunch to be triggered.
    • A "?" icon displays the specific code of available triggers and their meaning.
  • Ability to modify a relaunch item without having to delete and recreate it.

"Content items" feature:

  • The shortcode used to display a link to the list of content items (e.g. restaurants) in an FAQ response has changed structure to integrate customised filters and a customised pictogram. Structure to follow below:

    [showAllItems title="Link name visible on screen" picto='pictogram name.image extension' filterId='filter name' filterValue='filter value']

"Data sources" feature:

  • Ability to Deactivate/Reactivate a data source.

  • For PDF data sources, the "Download" button on the media card can be used to position the user on the page concerned by the requested subject. However, the PDF file type must be associated with a display application on your workstation to benefit from this positioning. For downloaded documents (Word, Excel, Powerpoint, etc.), this feature is not available and the document is displayed on its first page.

  • Ability to index "Content" data sources on Boolean fields: only the field label defined in the content type is displayed in the document partition selected by the semantic search.

  • Assign a "Tag" to a data source to filter the semantic search on this specific data source:

    • The website will pass this "Tag" to the Framework, which will then allow Retorik Studio to filter the data sources defined solely with this "Tag": in this way, semantic search and generative AI (if the function is activated) will only rely on the data sources associated with this tag to provide an answer.
    • Example use case: a single backoffice containing data sources from different entities (companies, managed assets, etc.) and different websites or different pages of the same website using the same backoffice: the website transmits the specific "Tag" to the Framework, which sends an event back to Retorik Studio, which filters on the specific data source.

    Event code:

{
  type: 'event',
  name: 'davi.setSemanticFilters',
  value: {
    tags: Array<string>
  }
}

Bug Fixes​

  • On mobile: sound and recording are now muted:
    • when the user switches to another browser tab,
    • when the phone is put on standby,
    • when the browser is minimised to return to the home screen.
  • "Hospitality" feature: the size of white pictograms in a "link to a list of content items" card (e.g. "list of restaurants") has been resized.
  • Multilingual: if the "Ignore welcome message" option is enabled in the Framework settings ("Chatbot > Settings > Integration" menu), the welcome message no longer launches when the language is changed.

Technical Update​

When generative AI is enabled, a streaming function can be activated in Retorik Studio ("Configuration > Settings > Chatbot" menu) to reduce the display time of the agent's response: the response is then displayed as the sentences received are processed, without waiting for the complete response to be generated.

Release Note March 25, 2024​

Retorik Studio: Version 1.14.0
Retorik Framework: Version 1.0.14
Retorik Kiosk: Version 1.1.0

New Features​

"Hospitality" feature:

  • "Discover" home card: the "Surprise me" button is now customisable by creating your own query on the content types and fields you want:
    • The display of the result remains unchanged: carousel of 3 random recommendations from the query result + a 4th card to see other recommendations.
    • Ability to rename the "Surprise me" button and the link on the 4th card.
    • These settings are available in the "Hospitality > Settings > Random recommendations" menu.

  • Content: ability to customise the title of the content items list:
    • Available in the "Content > Content Definition > Content Types" menu.
    • Select the content type, click the "Edit" button, fill in the "List mode title" field in the defined languages and "Save".
  • Content item description card:
    • New design of the Itineraries banner on the map in the bottom right.
    • Added a "Go" button redirecting to Google Maps displaying the selected transport mode on the banner. The user can then select one of the transport modes proposed on the web page (public transport for example).
      • On a Kiosk-type device, this button will display a QR code to scan to open the Google Maps route on mobile.

"Data sources" feature:

  • Ability to not display the data source in the agent's response:
    • This option is only useful if generative AI is enabled: the agent provides a response based on the checked data source, but does not display a media card in the second part of the response.

Framework settings: added configuration options from Retorik Studio ("Chatbot > Settings > Integration" menu):

  • Ability to hide voice mode and the "Sound" button to operate in text mode only (applicable to widget and full-screen formats). Selecting these checkboxes hides parts of the screen that become inapplicable.
  • For Widget format only:
    • Allow switching from widget to full-screen format, while retaining the current conversation.
    • In this case, a character size and a top-of-page margin must be specified for each format.
  • Speech synthesis:
    • Allow storage of speech synthesis data (option enabled by default when creating a Retorik Studio tenant)
      • Optimises the agent's response time by storing sentences already processed by the speech synthesis.
    • Ability to manage a list of words with their phonetic transcription (IPA code) so that the voice pronounces words correctly:
      • To be defined globally (for all languages) and/or language by language.

Kiosk-type devices:

  • Ability to launch the printing of a PDF document on a connected printer with a print preview. Feature available in:
    • The description card of a content item,
    • The download card of a PDF document associated with a knowledge base (FAQ) response,
    • And the download card of a data source.
  • The kiosk configuration is now retrieved from Retorik Studio at application launch and then at each wake from standby for elements that have not been modified via the kiosk configuration panel.

Bug Fixes​

  • Generative AI (Knowledge > ChatGPT feature): the number of document partitions sourcing the AI prompt is now the same as for the cards displayed by semantic search (data source indexing, Knowledge menu).

Release Note February 12, 2024​

Retorik Studio: Version 1.12.0
Retorik Framework: Version 1.0.12
Retorik Kiosk: Version 1.0.23

New Features​

  • Customisable "Discover" home card on the content types to highlight.
  • Customisable list mode on filters and search suggestions.
  • Semantic search:
    • On a file-type data source (text, PDF, Word, Excel, Powerpoint) and/or a content-type data source. Both types of sources can be combined.
    • A knowledge base question response remains a priority over the data source, which will appear in second position. Introductory text for the data source is customisable in the backoffice.
    • Ability to not trigger the semantic search with knowledge base responses.
  • Generation of text by an AI to populate a field of a content type (via the Workflow feature).
  • Added speech recognition configuration options in Retorik Studio settings:
    • Enable/disable a sound beep when the agent is listening. Beep disabled by default.
    • Adjust the silence duration before the microphone closes. Defined in milliseconds.
      Set to 1000 milliseconds by default. Do not exceed 10 seconds.
    • Word list: a list of specific words can be specified to improve their speech recognition.
      • To be defined globally (for all languages) and/or language by language.

Technical Update​

  • Integration of Orchard Core versions v1.7 and v1.8
    • Retorik Studio gets a new look! A cleaner interface supporting dark mode and light mode.
    • The context menu in the top right now explicitly shows the logged-in user.

Bug Fixes​

  • Two "undefined" buttons in the media confirmation popup in a content type.
  • The "Cancel" button on the "Select an image" page does not work.

Release Note January 22, 2024​

Retorik Framework: Version 1.0.11

New Features​

  • Visual loader when activating the microphone, to indicate to the user the actual listening phase of the agent and avoid the beginning of the question not being recorded.

  • Speech recognition configuration options:
    These settings are not yet available in the Framework settings accessible from Retorik Studio (Chatbot > Integration menu)

    • Enable/disable a sound beep when the agent is listening. Beep disabled by default.
    • Adjust the silence duration before the microphone closes. Defined in milliseconds.
      Set to 150 milliseconds by default. Do not exceed 10 seconds.
    • Word list: a list of specific words can be specified to improve their speech recognition.
      • To be defined in the configuration globally (for all languages) and/or language by language.

  • Enable/disable telemetry:

    • Telemetry logs application errors visible in the user's browser console. These logs are reserved for the DAVI technical support teams. Other technical information will be added soon.
    • Telemetry can be enabled in the Framework settings accessible from Retorik Studio (Chatbot > Integration menu). If these settings are not managed by Retorik Studio, telemetry must be enabled from the Framework configuration file integrated into the site.

Bug Fixes​

  • Issue with the display order of activities forming the agent's response (example: agent response followed by a list of documents).

Release Note November 27, 2023​

Retorik Studio: Version 1.11.0
Retorik Framework: Version 1.0.10

New Features​

  1. "Relaunches" feature:
    You've been waiting for it. Here it is! You can now relaunch the user after a period of inactivity with the digital agent. Refer to the Relaunches section for activation and configuration of this feature.

  2. "Hospitality" feature:

    • POI list:
      • On Desktop and Kiosk, the "Next" and "Previous" buttons have been replaced by infinite scroll. Once the last POIs are reached, a "Back to top" button appears.
      • On Smartphone or Widget integration, the "See more" button remains applicable.
    • Vertical Kiosk device:
      • Subtitles, previously placed below the cards and considered too low, now appear higher, in the position of the speech recognition area.
      • The "Language" tab has been reduced to harmonise its width with the control panel (for all devices) and, for the vertical kiosk, it has been moved below the control panel for better visibility.
    • Implementation of "Recipes" to initialise all intents related to the Hospitality, Weather and News modules, with a standard set of training phrases for the digital agent.
      • Note:
        • The "Davi.Retorik.Studio.News.Intents" recipe launches automatically when the "Davi.Retorik.Studio.News" feature is activated.
        • The "Davi.Retorik.Studio.Weather.Intents" recipe launches automatically when the "Davi.Retorik.Studio.Weather" feature is activated.
        • WARNING! The "Davi.Retorik.Studio.Hospitality.Intents" recipe must currently be launched manually.
        • These recipes can be manually relaunched on a tenant created before 27/11/23 (Configuration > Recipes menu): they retain the content of existing phrases, except for the 6 intents of the "Hospitality > Settings > POI Query" feature.
    • Multilingual management for content items associated with types: Food establishment, Lodging Business, Product, Service, Tourist Attraction and Tourist Destination (categories visible in the MENU for the Hospitality feature): it is possible for these content types to manually create content in a language other than French.
      • Note: Datatourisme/Décibelles Data data imports are not yet managed for multilingual.

  3. Default creation of the "ShowAllSettings" action for each Retorik Studio instance created, with a set of training phrases for digital agent learning. Reminder: this action is necessary to be able to use the Retorik Studio tenant with the Kiosk application on a kiosk or Windows touch tablet.

  4. Training:
    The previous version required FAQ training and NLP (Deep) engine training after creating a new FAQ or adding a question to an existing FAQ. To simplify the task for our Botmasters and avoid oversights, the [Train] button present in the "Knowledge > Knowledge Base" and "Knowledge > Dictionaries" features now executes both trainings automatically.

Note:

  • The end of FAQ training is notified by a "push" notification at the bottom of your screen, after a few seconds.
  • The end of NLP (Deep) engine training is notified in the notification centre, accessible from the "bell" icon in the title bar. Longer wait time (approximately 20 to 30 minutes if the Hospitality feature is activated with all intents).
  • Eventually, all notifications will be managed in the notification centre.

Technical Update​

  • Digital agent preview function integrated into the Retorik Studio backoffice, with the ability to share the site URL:
    • Accessible via the "Test" button (Robot icon) located in the top right of the screen.
    • A popup menu (right arrow) allows you to copy the exact URL to share with your contact.
    • The digital agent is integrated in full-screen mode.

Bug Fixes​

  • Card flickering in conversational mode (home cards, POI cards, question suggestion cards when the agent is uncertain, form cards).
  • Scroll issue in Widget mode: the website scrolled when scrolling up or down in the digital agent conversation window: this background appeared at the top on the line of the window close button, and at the bottom below the menu bar.

Release Note October 16, 2023​

Retorik Studio: Versions 1.10.1 to 1.10.7 (updates from October 2 to 9)
Retorik Framework: Version 1.0.9 (October 13)

New Features​

  • Choice of "voice/text mode": the conversation only starts after clicking the "Start" button.

Hospitality feature:

  • Home: The 4th card in the carousel launched from a category of the "Discover" card changes its behaviour:
    • It no longer launches 3 other recommendations but offers to display the list of all POIs in the category.
  • Animation added to the POI presentation:
    • The POI list appears from the left,
    • The POI card appears from the left,
    • The card carousel arrives from the right.
  • Product card:
    • The phone number is moved out of the button,
    • Photos: zoom is applied to vertical and horizontal photos so they fill the entire landscape area.

Technical Updates​

  • Digital assistant integration settings from Retorik Studio:
    • Prerequisites:
      • Website side: the "config" field of the properties used when instantiating the main component must not be filled in, or must have an undefined value.
      • Enable the "Davi.Retorik.Studio.Integration" feature in Retorik Studio: the "Integration" function must appear in the "Chatbot > Settings" menu.
    • Available settings:
      • General settings: integration parameters, default voice or text mode, character selection, voice selection, etc.,
      • Image settings: logos, home background and news ticker background,
      • Colour settings: customisation of button colours, media cards, texts for standard features as well as "Hospitality" and "Forms" features if they are activated.
  • Browser sound caching to reduce Azure TTS (Text To Speech) service costs.
  • Change of the default font from "Helvetica Neue" to "Calibri sans-serif" to fix character height differences on touch kiosk devices that made reading difficult.
    • Note: for iOS and Mac devices, if the Calibri font is not installed, it will automatically be replaced by the "Helvetica" font.

Bug Fixes​

  • Hospitality feature:
    • POI list in voice mode: removal of repeated page title (e.g. "All places to eat") when filtering or clicking the "Next" button.
    • POI card: the colour code of the card downloaded in PDF format does not match the POI category.
    • Content items: deleting an item corresponding to a POI category in Retorik Studio does not remove the POI from the list in the digital assistant interface.
  • Multilingual:
    • Knowledge base: import of responses in a foreign language does not work.
    • Interactive map not translated (excluding POI content).
    • Translation not possible in Retorik Studio for foreign languages defined in the "Configuration > Settings > Localisation > Cultures" page for:
      • The 3 featured questions after the welcome message,
      • "Newspost" type content items (news ticker),
      • Weather intents in the "Weather > Settings > Intents" menu,
      • News intents in the "Chatbot > Settings > Intents" menu,
      • All intents related to POI categories in the "Hospitality > Intents" menu,
      • The 6 POI intents on the "Hospitality > Settings > POI Query" page.
  • iOS: overhaul of the voice management system for iOS, to fix voice dropouts in the previous version.
  • Text mode: when a conversation remains open for a long time, the system manages to pick up voice input.
  • Animation engine: video sizes have been readjusted to reduce graphics memory usage and eliminate unexpected crashes encountered in the previous version.
  • Standby on Kiosk-type device: each time the news ticker launched, a conversation was generated, altering conversation statistics. With the fix, the news ticker runs in the same conversation which renews every 7 days.