Changelog

December 23, 2024

The AI no longer switches languages automatically when asked a question in a different language. The user will instead be prompted with the question whether they would like the conversation language to be changed and they can select between Yes/No. If they click yes, the conversation from that point on will be in said other language.
We've improved how the AI searches through the available sources to find the most relevant chunks to answer a question.

December 16, 2024

You can now limit the language(s) the AI is allowed to use via the Configuration tab. By default, all languages are allowed.
You can ingest more content types from Salesforce (not just articles), using custom query.
We’ve improved how we retrain sources. This lays the groundwork for several upcoming features so stay tuned.
We’ve improved the translation corpus.
With a redirect action set to open in a new tab, the box would appear unchecked even though it would open in a new tab as intended. This has been fixed.

December 9, 2024

We've added support for links & attachments to the Salesforce livechat.
We've improved rendering in the Salesforce app so the text appears as formatted.
Copy to feature for components works again.

December 2, 2024

With AI actions, you could already collect information from users via text input (eg. email address). Now you can also offer them multiple options to choose from by clicking to select.
We thoroughly tested the browser assistant which resulted in a lot of stability improvements and bug fixes.
We’ve fixed bugs related to insights when switching environments, PDF ingests, uploading multiple FAQs, and generating candidate questions.

November 25, 2024

Updates to the team assistant:

You can now see the previous conversation history when re-opening a ticket. The AI will not generate a new response from scratch again.
We’ve also added the option to Reset a conversation if needed.
Notes are now available across sessions. To reflect this change, the Create note button has been renamed Notes.

November 18, 2024

The browser assistant can now summarize and answer multiple questions per ticket.
You can now filter conversations based on the segment.
We've implemented a better PDF parsing library. If you have indexed PDFs, you can retrain them with the improved system to see if the results are better.
We've adjusted the AI to make less arbitrary decisions. When multiple articles could answer a question, the AI now summarizes options and asks for clarification before proceeding, ensuring more relevant responses.

November 11, 2024

You can now review and accept or reject the candidate questions generated by Unless based on your historical tickets. When accepted, they will be added to your quality control questions. (Please contact us if you’d like your top 20 questions to be generated based on tickets.)
To accept a candidate question, you have to write a control answer first. We know this can be tricky sometimes so we’ve added a Generate answer button you can use to prompt the AI. You can use the generated answer directly or have it be a starting point and make edits as needed.
You can now filter conversations based on email address.
The Freshdesk integration now supports multiple portals.
The note feature in the browser asisstant works again.

November 4, 2024

New AI view: the side panel 🎉 Until now the AI could be set to open as a popover (a small rectangle, often on the corner of a screen) or as the larger conversation center overlay. There’s now a third option! You can select the side panel view which will slide to open and close the chat.
You can now add a search bar component which will open up the larger conversation center overlay where the end-user can interact with the AI to find the information they are looking for.
We improved language detection stability.

October 28, 2024

The AI now uses more of the conversation history for context when providing answers.
We no longer show single-word inputs in the AI zone insights (eg. hello, okay, bye).
There is now a setting to remove the Search tab.
The contact label will also be used for the title of the contact page.
You can now bulk upload multiple PDFs and markdown files. Each row in the file will be imported as a resource, all grouped under a single training source.
When uploading a CSV file, we now check the header columns for "title" and "description" to use for the title and content.
Content Security Policy (CSP) is a security feature that helps prevent a range of attacks such as cross-site scripting (XSS) and data injection by specifying which sources of content are considered safe. If you are using a CSP then you will need to add the Unless domains to the CSP for it to work. You can read more about this in the Unless Developer Hub.

October 21, 2024

Weekly account summaries are now live in Slack! 🎉
We improved the startup load time of the AI component and it now loads much faster after clicking the floating button for the first time.
The starter message when using the browser assistant will no longer be translated and show in the original language it was written in.
Bug fix regarding AI zone insights to refresh automatically

October 14, 2024

You can now trigger an action (eg. contact support) after an unanswered question as well as after every single response.
You can add a javascript rule to determine if/when an action is shown. For example, you can set up a “live chat handover” action which will only get triggered during the week, within business hours.
A download icon has been added to the chat window, allowing the user to download a chat transcript.
Live chat handover to Salesforce is now possible! 🎉
The Salesforce integration for ingesting sources now supports the Body field.
In the conversations list view, you can now add a column to see the current URL and filter based on it.
The resource list now has a column for publish date (if available) which can be used to sort the list and prioritize your content efforts.
You can add a pre-action message to show before asking for user input. For example, if asking for an email address, you can add a note that says “You can leave your email address here and our team will be in touch shortly”.
Fixed a UI issue with the chat incentive buttons, improving consistency.
Fixed a bug regarding the email collection action.

October 7, 2024

When adding a training source, it is now possible to bulk import via CSV. Each row in the file will be imported as a resource, all grouped under a single training source.
You can now edit an existing environment to be a test environment and vice versa.
Fixed an issue where the AI component wouldn't scroll if actions were added after a thumbs down.

September 30, 2024

⚙️ Wordpress knowledge base integration: You can pull your Wordpress pages, posts and articles into Unless to use as training sources for your AI.

September 23, 2024

You can now configure any action to be displayed to the user after negative feedback. This means instead of just having an email support flow, you have more options like redirecting to forms, call-me-back, book a meeting, live chat handover, etc.
PII will be filtered by default from the user input when having a conversation with the AI. (This is configurable.)
There is now a disclaimer at the beginning of AI conversations to inform the end-user they are talking to an AI in line with the transparency obligations of the EU AI Act. (This is a message you can edit and customize.)
API AI actions are here. 🎉
You can now import Markdown files.
When a response from the AI is too long to fit in the view, it will no longer scroll to the bottom of the message but show the beginning instead. A ‘scroll to bottom’ icon shows in these instances or the user can manually scroll to the end of the response.
Admins can now delete questions from “Most asked questions” and “unanswered questions” lists in Insights. Sometimes these lists include test questions that you might want to remove, otherwise they’ll be on the list for 30 days.
Bug fix: Conversation details no longer disappear when jumping to ‘read’ status.

September 16, 2024

Chat incentives are here. 🎉 You can now add action buttons to the floating starter button which will already be visible before the user opens the chat and starts interacting with the AI. By predicting the questions of users and guiding them to the answers they might need, we can increase engagement and satisfaction.
You can now generate “candidate questions” for the quality control report based on historical tickets.
Admins can now delete a conversation (ie. if it includes PII). A record of the conversation will be kept (so that Conversation URLs aren’t broken) but the chat data will be removed and replaced with a message that the conversation has been deleted.
Fixed bugs related to segment options not showing in some new conversations and the selection being lost when a chat is closed and re-opened.

September 9, 2024

Conversations list view filters are remembered. This means if you filtered the conversations, go to the details page of a conversation and go back to the list, you will still see the filtered view.
Similarly, if you’re viewing the details page of a conversation and click the Next conversation button, it will now take you to the next conversation from the filtered list.
“Searching...” and “New Conversation 1" fields are now editable so if you prefer different wording or a different language, you can overwrite the default text.

September 2, 2024

It’s now possible to view conversations that are older than 30 days.
Font selection is now applied to AI components as well.
The AI editor has new fields (some of which you can see in the screenshot below) where you can overwrite the text of buttons, labels, and placeholders.
You can use the hide app bar setting in the AI editor to only show the chat tab and hide the search and contact tabs.
The AI settings have been adjusted to make it more strict and less creative in it’s answers.
You can now create a translation corpus to force the AI to use certain translations that are not so common.

August 26, 2024

All accounts have been upgraded to a better model.
You can filter the quality control report based on score.
You can integrate the browser assistant with Freshdesk and configure your groups to environments and (optionally) segments in Unless.
Segment names are now shown in the conversation details and the quality control report per question/answer.
When redirecting to a URL (as an action), you can now set it to open in a new tab.
Messages are now grouped and separated per segment. This means when in a segment, you will only see the messages in that segment and if you switch to a different segment, you will only see the messages in that newly selected segment.
Bug fixes

July 29, 2024

You can now review your quality scores over time in a graph to track progress more easily.
When viewing a resource, you will now see the publish date (if available), allowing you to prioritize your improvement efforts based on this data point.
Tables now render better when viewing a resource.
Switching pages while in the search tab of the browser assistant now also triggers a new search based on the new ticket/page.

July 22, 2024

UI improvements to the QCC report to enhance readability
You can now filter your sources based on quality score to troubleshoot more easily.
Add commonly unanswered questions to the quality control center at the click of a button.

July 13, 2024

Defining pages to include/exclude when ingesting a website just got easier.
You can now exclude a certain page from training data with a simple click.
⚙️ Freshdesk knowledge base and ticket integrations are now live. 🎉
You can copy an AI response with a single click using the "copy to clipboard" icon that’ll appear when you hover over a response.
Quality control questions are numbered (also in reports) and remain in the same order making it easier to refer to specific questions.
Bug fixes regarding AI responses and retraining of sources

July 8, 2024

🎉 Live chat handover (to Salesforce, Zendesk, etc.) is now possible inside an Unless AI component! You can add this as one of the options in the contact tab. Additionally, you can add an FAQ with an action to trigger this when a user requests to “talk to a human".
Bug fixes regarding the injection of formatted content such as bullet points

July 1, 2024

It is now possible to exclude CSS selectors when ingesting a website. This enables you to only ingest the content you need while avoiding the clutter around it (like a table of contents, breadcrumbs, etc.).
You can now ingest data from the Visma Communities website using the Visma Data Lake.
Content shown in the search tab now includes a page description (if available) in addition to the page title and URL.
You can now see the number of resources you have in global and in each segment at a quick glance on the Sources page.
Bug fixes and UI improvements

June 24, 2024

There's now a search tab that can be used to look up similar content to the question that was asked. If no question was asked yet, the AI will find content similar to the page that the user is on. The search tab can also be used to find similar tickets via the browser assistant.
You can also add a contact tab to your AI components so that your users can get in touch directly when needed.
You can now review commonly unanswered questions & most downvoted questions at a glance via the Insights tab of your AI zone.
New settings have been added to the browser assistant configuration. It is now possible to add a logo, a starting message, and a placeholder.
Global sources and segments are now combined under one tab: Sources.
UI improvements have been made to the quality control report.
We’ve added support for GPT-4o.

June 17, 2024

You can now see a breakdown of the AI friendliness score for your sources. We’ve added an Improve your sources section to the following article where you can find out more.
🎉 Freshdesk integration: enabling you to automatically get answers from the AI, using the Unless browser assistant.
You can now set a default language for the browser assistant on this page by clicking Configure and then Settings.
Setting a default conversation name is now possible.
URL mentions are filtered before content is scored on AI friendliness for more accurate ratings.

June 10, 2024

It is now possible to add a note to a conversation in the browser assistant as well.
AI friendliness score now includes a metric for minimum text length.
AI friendliness sort now works as expected.
If an audience is stopped, it will still be possible to see its label in the experiences it’s attached to.

June 3, 2024

Users will be identified when using the Test your AI feature in the dashboard so that it’s possible to show them in the conversations list and details pages.
Segments now have a type so it’s easier to distinguish content vs tickets.
You can now automatically show the AI popup (after a delay) instead of waiting for the end user to click to open it.
The AI friendliness score now includes a sentence analysis. This means measuring how many sentences there are in a chunk. If the percentage is low, this indicates bad content (like an HTML footer being ingested).
We now automatically clean audiences that haven’t been used in some time.

May 27, 2024

When configuring the browser assistant, you can connect Zendesk brands or Hubspot pipelines to segments. The browser assistant will also detect which of the brands/pipelines you're currently viewing and it will automatically switch environments accordingly.
When viewing the source(s) used for an AI response, the end-user will only see the sources that have been selected by the AI. In the Unless dashboard, however, you will now be able to see all related sources as well.
User roles in Unless have been changed to administrator, user, and browser assistant only.
Messages in a conversation were sometimes showing out of order. This has been fixed.
UI improvements were made to the browser assistant.

May 20, 2024

When reviewing your indexed training sources, you will see an additional column for AI Friendliness. This is a new feature that is currently based on the length of a source however more factors will be added, along with related features.
FilterPII is now set by default on the summarization API.
You can now select this quarter and last quarter as preset date ranges when viewing AI insights.
The browser assistant is now available in the Edge add-on store (in addition to the Chrome web store which we recommend using instead).
You can check the sources used for an answer in the quality control report as well.

May 13, 2024

It is now possible to join an audience when viewing a component.
When an integration fails to sync, you will see an error in the dashboard.

May 6, 2024

AI action variables are here! 🎉

You can now use variables when adding an action. Think of variables as something that needs to be known for an action to be completed. For example, for actions related to an Order ID, the AI needs to know the order ID. If it isn't already known, the AI will first ask the user for input before initiating the action.

Setting up an action with a variable

Currently there are two options available: string and regex. We have more variable types on the roadmap but if you have any specific requests, please get in touch!

The browser assistant will now set "Email" metadata to reflect which tickets were handled by which agents.
Quality control questions are no longer factored in when showing most commonly asked questions and top sources.

April 29, 2024

⚙️ Azure Devops WIKI integration: You can now pull your WIKI documents into Unless to use as training sources for your AI.
We’ve updated the browser assistant with a limited host list for increased security along with UI improvements.

April 22, 2024

🎉 The Unless browser assistant is now available in the Chrome store! We will share more details about this soon.
In highly requested features, the AI will now start in the browser language by default. Using the editor, you can change this setting by choosing one of the options from the dropdown. Additionally, AI actions will also be translated to the language of the conversation.
While in the editor, you can also select starting actions from a dropdown to follow the starting message of the AI.
You can now see which actions are attached to which FAQs at a quick glance, under training sources.
Reviewing conversations just got quicker with the Next Conversation button.
You can now see a URL of the page where the conversation took place whether it be on your website, product dashboard, in your ticketing system like Zendesk, Salesforce, etc.
We released a new version of the AI component which allows us to more easily develop new features on top of it. You shouldn’t notice any differences but if you do, please let us know. This update will enable features like AI actions, text-to-actions, and more to come.
When a prompt is filtered by policy, a readable error message will be returrned to the user.
As always we also fixed some bugs and made improvements to the UI.

April 8, 2024

New: Conversation status 🎉

Keeping track of conversations just got easier!

From now on each conversation will have a status, starting with New and switching to Read once you've opened up the details page for that conversation.

Most importantly, you can see which responses have been flagged since the conversations list can be filtered for the Flagged status and in the details view the flag icon will have a red outline.

If the reason for the flagging is that the AI didn't have the answer or gave a wrong answer but the issue has been resolved (eg. by adding or updating sources), you can then change the status to Closed. This makes it easier for your team to be on the same page and direct their attention to where it's needed.

Show conversation status

More functionality when flagging a response 🚩

It is now possible to flag repsonses as part of your internal evaluation and not just for reporting issues to the Unless team. This way, when analyzing conversations, you can flag responses and leave notes for your colleagues describing what the issue is and/or what the answer should have been.

Reporting a flagged response to the Unless team as a technical bug is now optional.

At this stage, you can also add the question to your Quality control center. Don't worry if the question isn't how you would structure it, once you click save, you will have the possibility to edit the question and the control answer.

Update question in the quality control list

March 18, 2024

🏆 Most common questions in the last 30 days

In the AI zone, you can now see a list of the top 10 questions that were asked to your AI in the last 30 days. With this information, you know where to dedicate your attention to ensure the AI can answer these questions.

Most common questions in the last 30 days

🏅 Most common used sources in the last 30 days

In addition to the most common sources, you can now also see the top 10 sources most commonly used by the AI when answering questions. Ensuring that these sources are properly indexed, accurate, and up-to-date will provide better results.

Most common used sources in the last 30 days

You can read more about these two features here or better yet go into the AI zone and find out what your top questions and sources are! 👀

You can now filter the conversations list by conversation name.
You can now delete FAQ entries using the CSV upload. This is particularly helpful for bulk deleting.
Training source edit view has a new field for the segment that a source should (optionally) belong to.
Bug fixes and UI improvements

March 11, 2024

Insights are no longer cached if the selected date is in the future.
Links in the meta-data block of the conversation details view are now clickable.
Sitemap xml files with relative paths are now supported.
The billing page now also shows environment and segment usage.
🤫 We now have a beta version of the Unless browser assistant.

February 26, 2024

See audience insights when selecting an audience from the audiences list, similar to how it is with components.
It is now possible to use full URLs with domain name when previewing components.
If you are using Hubspot, your support team members can now leave a comment as feedback on conversations.
Bug fixes and UI improvements

February 5, 2024

It is now possible to join an audience when closing a component. This way you can be certain that the viewer has seen the message and you can opt for showing a different component next time.
When checking the sources used in a response while at the conversation detail view, you now see an edit button making it easier to quickly make updates to a source when applicable.
When the end-user selects an option during a conversation (eg. a follow-up question or an action) the options provided will no longer disappear.

Prioritization and exclusive experiences

The prioritization tab has been removed but you can still prioritize your experiences under advanced settings.

The priority number determines when to load the experience compared to the other eligible experiences. Priority 1 is the highest. Priority 25 the lowest. When viewing the experiences list, you can see the prioritization at a glance and change the priority order on the spot.

Additionally, you can now set exclusive experiences. When setting the experience to exclusive, it will be the only experience that can be shown to the visitor. Exclusive experiences are always shown first.

Experiences list view with prioritized and exclusive experiences at a glance

January 29, 2024

AI Actions are here! 🎉

Actions can be attached to FAQs to activate and engage your end-users and get them to take the next step, leading to more actionable interactions.

With actions, when your selected FAQ is asked, the end-user is given options like starting the support flow, getting redirected to a URL, seeing a component and more, along with an answer from the AI.

Check out this help center article to find out how you can answer with AI actions!

Actions seen live in a chat component

January 22, 2024

We've fixed bugs and made improvements to the UI of the AI chat.
🔥 Soon you will be able to attach and initiate a variety of AI actions via conversations, resulting in more actionable interactions.

January 15, 2024

We've fixed bugs and improved AI stability for an even smoother experience.

January 8, 2024

🙅 Filter personal identifiable information (PII) when adding training sources. We then filter personal information like names, personal IDs, email addresses and more from the source before ingesting.
🎭 You can now define your own custom persona via the Configuration tab of your AI zone. You can hover over the ? icon next to the other persona options to get an idea of the kind of input you can provide. Keep in mind that you can use a maximum of 200 characters.
💬 The conversation popover and conversation center are now merged. Enlarging the chat popover will bring users to the conversation center where there is an overview of recent conversations. Clicking the size icon again will bring the user back to the popover view.
🎨 You can open up the editor to customize your AI component via the Configuration tab. While in the editor, you can also switch between different components that are on the same page.
🧩 Reorganized the component templates list into new categories for better user experience and ease of finding what you need.
🎟️ Similar ticket functionality is now available as part of the Hubspot and Salesforce integrations. In the case of Hubspot, it is also possible to filter by pipeline.
⚙️ Clickhelp integration: It is now possible to pull your Clickhelp knowledge base articles into Unless and use them as training sources for your AI.
📱 On mobile, the chat popovers display as overlays for better usability.
Large PDF files are now handled better.

December 18, 2023

Conversation details pages now show responses in markdown so if the response originally contained bullet points, they will appear as such, URLs will be clickable, etc.
👍 You can now add a reply for when a user gives positive feedback to the answer provided by the AI.
⚙️ Confluence integration: It is now possible to directly pull documents from Confluence and use them as training material for your AI.
We’ve renamed the Websites tab to Environments. Many customers are using Unless in their CRMs, support environments, etc. so we wanted to make it clearer in the dashboard. Your website is an environment but you can use the Unless AI in other environments too.
Conversations are now readily accessible from the menu on the left.
Fixed some bugs and improved the UI.

December 11, 2023

We've added new AI chat button templates which are highlighted in the screenshot below. Using these templates opens up more editing capabilities such as setting an icon, adjusting the color and size, etc.
Fixed some bugs and improved the availability of the AI as well.

December 4, 2023

🌟 The AI now responds 30% faster and handles follow-up questions better.
📚 You can now see exactly which pages/content was ingested from a training source. This helps you see which content was discovered using your sitemap, or which articles were retrieved by an integration.
You can click on each listed item to see the exact content that was ingested in Unless, helping you debug any problems.
📋 Additionally, now when viewing conversation details, you can click to see the sources and click again to view the exact content that was used to generate each answer!
ℹ️ The sources list (that can be viewed for each answer) is now much more accurate and will show exactly which sources were used in the answer.
👍 The thumbs up and down icons (and the sources) are made more visible and appear below the response to increase usage.
📞 An (optional) support button is added with a new support flow in the top-right corner of the chat. You can set this up via the Configuration tab.
📌 We’ve added an indicator to highlight the last conversation you opened to make it easier to know where you left off while browsing across conversations.

November 20, 2023

The subject line for emails delivered to your specified customer service email is now customizable via the Configuration tab of the AI zone.
You can now adjust the website names listed at the top right corner.
Added article visibility properties to the Salesforce integration such as include/exclude options.
We’ve made some changes to make the input field in the chat clearer.

Language detection

We now store the determined language of each message and answer which should make it easier to identify outliers and errors.
Additionally, you can add a fallback language for cases where the language cannot be confidently determined. You can do this via the Configuration tab of the AI zone.

Fixes

The chat no longer crashes in the case of an empty message.

November 13, 2023

In Conversation detail pages, you can now flag a response to report an issue to our team. It is also possible to provide additional context here in terms of what happened and what you would expect to happen so we can address the issue accordingly.
If you have multiple AI segments, you can ask the end-users to self-segment at the start of a conversation.
In the Billing section, we added graphs to make it easier for you to review the use of AI responses over time.
We added breadcrumbs to our dashboard pages to make navigation easier.
You can now exclude an article ID when using the Zendesk integration.

November 6, 2023

🎉 Changelog: From now on we will use this page to share what we are working on so you can stay up to date on all new features, improvements, and fixes.
⚙️ Zenya integration: It is now possible to directly pull documents from Zenya and use them as training material for your AI.
The height and width of the AI chat pop-up is now adjustable via advanced settings in the editor.
You can now add your sitemap as a training source using a URL or by uploading a sitemap xml file.

October 30, 2023

It is now possible to rename your training sources, making it easier to browse and find what you’re looking for.
Zendesk multibrand is now supported.
We can now ingest tickets from Zendesk.
Subcategories can now be fecthed with the Salesforce integration.
Conversation sentiment was calculated based on sentiment analysis of the questions and the ratio of positive and negative responses from the user. This is now split so you see the sentiment and rating separately. Fixes
When reviewing conversations, if you were going between the list view and the detail pages, your filters would reset each time. This is now fixed.
The configuration wasn’t loading when switching websites.