New features and improvements

• Everything in one list: on the left every function with "On" or "Off", on the right the settings of the one you pick. Webhook and HubSpot now live under "Pass on automatically".
• Appointments right in the chat: free times show up below the answer, sorted by day. Tap a day, pick a time, fill in a short form, booked. Whatever the visitor already wrote in the chat, such as their name, is already in the form. Up to 5 appointment types, for example a free 30-minute consultation and a 1-hour demo. The visitor gets a confirmation with a calendar file.
• Contact details your way: name, email, phone and request each as Required, If possible or Don't ask, plus up to 5 questions of your own, such as company. If the chatbot does not know an answer, it offers to pass the question on to your team.
• Live chat with hours: set per weekday when someone from your team is available. Outside those hours the chatbot promises no one and takes the request instead. Live chat now also works in the inline chat embedded in your page.
• Passed on automatically: new requests and bookings land in HubSpot or at your webhook. From the Standard plan.
• Where: Assistant → Functions

• Suggestions from your website: the assistant reads the pages your chatbot already knows, or takes a quick look at up to 20 pages. It also checks what is in your main menu, on which pages your visitors ask the most in the chat and what they want to know there, and starts right there: start page, pricing, key product pages. Large websites get up to 10 suggestions. Each card shows how the message will look in the chat. Example for a bike model's page: "Between two frame sizes? Tell me your height and I'll say which one fits." Tick the ones you like and click Use selected
• Every language version: a message for "/pricing" also appears on "/de/preise" and the page's other language versions, as long as your website links them. Visitors see the message and answer buttons in their own language while Automatic translation is on. You can switch this off per message with Also on the page's other language versions
• Answer buttons: visitors see up to two one-click answers under the message, such as "Compare plans", so nobody has to type
• Goal for the AI (Premium): one sentence on what the conversation should achieve, for example "Ask about the team size and recommend the right plan". The chatbot works toward it as soon as a visitor answers
• Rarely answered: when a message is shown often but hardly anyone replies, the editor marks it and offers Suggest a new text
• Where: Dashboard → your assistant → Channels → Chat widget → Automatic engagement → Get suggestions. New chatbots also offer it in the last setup step. Available from the Standard plan

• Error messages in the visitor's language: when a reply fails, visitors now read the notice in German, French or any other chat language instead of English
• Clear hint on the wrong website: if the chat runs on a website you have not allowed, it now says what to do: add the address under "Allowed websites"
• No long dashes: the chatbot writes with commas and colons instead of long dashes, which reads more natural
• Set the form of address: in languages such as German or French, add one sentence to the answer style, for example "Always address visitors formally". The chatbot then keeps it for the whole conversation
• Where: your assistant → Settings → Answer style and role

• Your choice per person: when you invite or edit a team member, tick Two-step login required. It is off by default
• Set up first: after the next login the person sees only the setup. They connect an authenticator app and save their recovery codes, then the dashboard opens
• Stays on: the person can switch two-step login off again only after you remove the tick
• Where: Settings → Account → Team members → Invite or Edit

• Same colors as the chat widget: besides the theme color, set the text color on the theme color plus background and text color for bot answers and visitor messages
• Automatic until you change it: empty fields follow your theme color, so existing inline chats look the same. The arrow next to a field resets it
• Readable by default: pick a dark bubble and the text turns white on its own. If a combination is hard to read, the settings show the contrast ratio
• Fixed: with a very light theme color such as white, the send button and visitor messages now use white text instead of dark grey on grey
• Where: Dashboard → your assistant → Channels → Inline chat → Appearance

• Videos from your website play in the chat: when a video on your website fits the question, the chatbot shares it and a video card with a preview image takes the place of the link. Several videos appear as a row of cards. Works with YouTube, Vimeo and MP4/WebM files
• Privacy built in: YouTube and Vimeo only load after the visitor clicks play. Until then nothing is sent to YouTube or Vimeo; the preview image comes from our server. YouTube runs in its privacy-enhanced mode
• Images edge to edge: a single picture fills the chat bubble like in a messenger, several appear as a row. A click opens them full screen, Esc or the close button brings the visitor back
• Products as a row: product recommendations with picture, name and price appear as a row visitors can swipe, drag with the mouse, scroll with the mouse wheel or browse with the arrow keys. Now in the inline chat too
• Accessible: everything works with the keyboard and screen readers, and the price colour follows your brand colour only where it stays readable
• Your own CSS: new stable classes for custom CSS: .wca-video, .wca-videos, .wca-image, .wca-gallery, .wca-image-viewer (with ::backdrop), .wca-products and .wca-product-card
• Videos on by default: switch off Play videos in the chat under Dashboard → Integration → Chat widget → Behavior or Integration → Inline chat → Content & Behavior, and answers show a plain link instead
• Existing knowledge: videos are picked up the next time your website is re-indexed, automatically at your re-index interval or right away with the refresh button (Re-index) on your website under Knowledge

• Extra code at sign-in: after your password, or after signing in with Google, you enter a 6-digit code from an authenticator app on your phone, for example Google Authenticator or Microsoft Authenticator
• Your password alone is no longer enough: someone who only knows it can't get into your account
• Recovery codes: you get 10 one-time codes for the day your phone isn't at hand
• Team members too: every team member can turn it on for their own login
• Optional: it stays off until you turn it on
• Where: Settings → Account → Two-step login. Step-by-step guide in the docs under Account → Settings

• Your design, your app: developers can build their own chat window with the free WebChatAgent SDK (npm package "webchatagent") instead of using the ready-made widget. Same chatbot, same knowledge, same settings
• Everything the widget does: answers that appear word by word, suggested follow-up questions, sources, file upload, thumbs up/down, lead forms and handover to your team with live chat
• Runs everywhere: in the browser, in Node.js and with React, Vue or any other framework. Open source under the MIT license
• Ready for AI coding assistants: the guide has a section for tools like Claude Code, Cursor or Copilot, so they set up the chat correctly
• Where: npm package "webchatagent", guide in the docs under Developer Guide → JavaScript SDK, overview on the Chatbot API page

• Chat on every page: install the WebChatAgent extension, paste your chatbot ID into the site configuration (tab Chat) and the chat button shows on all pages of that site. No TypoScript, no template changes
• One chatbot per site: TYPO3 installations with several websites set a different chatbot ID for each site
• Hide it where it doesn't fit: a switch in the page properties hides the chat button on a page and all its subpages
• Inline chat element: put the chat inside a contact or help page with the content element Inline chat
• Logged-in website users: optionally hand their name and email address to the chat, so the bot doesn't ask again
• Works with TYPO3 10 to 14: one release for all versions, install via Composer or the Extension Manager
• Where: TYPO3 Extension Repository, search for "webchatagent". Setup guide in the docs under Integrations → TYPO3

• One conversation: open it and click Export at the bottom of the window. Handy when you want to forward a chat to a colleague in sales
• One conversation as a PDF: in the same window, Export → PDF (print or save) opens your browser's print window, where you can print the chat or save it as a PDF
• Only the ticked ones: tick conversations in the list and click Export above it. The button shows how many you picked, e.g. Export (2)
• Everything, as before: with nothing ticked, Export downloads all conversations that match your current chatbot, search and date filters
• Same file format: a CSV with one row per message that opens in Excel, Numbers or Google Sheets. Available on every plan
• Where: Dashboard → Conversations

• Instagram direct messages: connect a professional Instagram account (Business or Creator) and your assistant answers incoming DMs with the same knowledge, instructions and tools as on your website
• Facebook Messenger: connect a Facebook Page and your assistant answers its Messenger messages. If you manage several Pages, you choose the one for this chatbot
• Comments under your posts: new comments on your Instagram posts and Facebook Page posts get an answer. You pick the mode: reply publicly below the comment, or leave a short note there and continue in a private message (recommended for product questions)
• Collects what your team needs: product interest, callback requests and other details you ask for end up in your Leads
• Your team can take over: once live chat is requested or open, new messages go to your team and the AI stops answering in that conversation
• Open about the AI: the first private reply says it comes from an AI, and the first AI reply in a comment thread is labelled
• No Meta developer setup: sign in with Instagram or Facebook from the dashboard, the connection runs through WebChatAgent
• Current limits: automatic replies handle text; messages with only an image or file get a short notice. One Instagram account or Facebook Page per chatbot. Available on Premium and higher plans and during a trial
• Where: Dashboard → your chatbot → Channels → Instagram or Facebook Messenger

• No more model hunting: choose "Auto (EU) – recommended" as provider and your assistant always runs on the model we currently rate as the best fast one
• Two fixed rules: Auto only uses models that count 1× against your message allowance and process data inside the EU
• Today that is Claude Haiku 5.5 on Google Vertex AI in the EU. When a better model meets both rules, we switch Auto over and you don't have to do anything
• New chatbots start on Auto. Existing chatbots keep their current model until you change it
• Where: Dashboard → your chatbot → Configuration → AI Model (Provider: Auto (EU) – recommended)

• Phone leads go to your team: when a visitor left a phone number, the notification only went to your account email, even if you had entered other recipients. It now goes to the addresses you set for leads, just like email leads
• Forms in any language: when visitors fill in your Action Bar form in Russian, Spanish or another language, name, email and phone are now recognised. Before, a Russian lead could be saved without its phone number
• Lead details in your language: field names and dates in the lead view now follow your dashboard language
• Where: Dashboard → your chatbot → Tools → Leads (recipients), and sidebar → Leads

• Space bar works on every website: on some WordPress themes (for example Customizr), visitors couldn't type spaces or use the arrow keys in the chat window because the website's own scripts grabbed those keys. The chat window now keeps them
• Italian visitors get Italian: greetings, prompt buttons and messages sometimes stayed in English for Italian visitors, and live chat messages reached you untranslated. Fixed, also for a few other languages
• Embedded chat keeps its shape: on pages without a fixed height it no longer grows endlessly, and the suggestion buttons no longer stretch into huge pills
• Connectors with non-Latin names: a connector named only in Cyrillic, like "Статус заказа", made every chat of that bot fail with "A connection error occurred". Fixed
• Umlauts in the email assistant: some mails showed "Grüße" instead of "Grüße". Fixed
• Where: works automatically, nothing to change

• Stops floods early: set how many mails from one address the assistant answers within 24 hours. Default 10, up to 50. Anything above gets no draft and no reply and stays untouched in your mailbox
• Contact forms: if your form sends every enquiry from the same address, raise the value
• Microsoft 365 and Outlook.com: the help pages now explain signing in with Microsoft 365, including shared mailboxes
• Where: your chatbot → Channels → Email → Max. mails per sender per day

• Straight from the mail: open a mail in the Email area and choose More options → Stop answering mails like this. Pick this sender, the whole domain or a piece of text in the subject. The assistant's draft for that mail is deleted right away
• Stops notifications at their real sender: a sender filter checks the sender and the reply-to address, so a no-reply notification that names your customer as reply-to address no longer turns into a draft for that customer
• Easy to undo: every filter is listed under Mails without a reply, where you add new ones or remove old ones. Matching mails get neither a draft nor an automatic reply, and they stay in your mailbox
• Where: Email area (More options on a mail) and your chatbot → Channels → Email → Mails without a reply

The new Email area brings your mailbox into WebChatAgent. Open incoming and sent messages, write emails and manage drafts directly in the dashboard, with an AI assistant alongside your work. It connects to your existing email account.

• Your emails in one place: switch between connected mailboxes, open folders and continue drafts from your mail program. Each email and draft has a direct link you can copy and reopen later
• An AI conversation beside your draft: ask "Have I already answered this?" or say "Make the reply shorter" in the same field. The assistant can answer, edit the draft or ask a follow-up question. It remembers the conversation and answers in the language you use
• See what was said before: a timeline below the draft shows the email history. Open cited messages to check the original wording before replying
• Write with fewer steps: choose known recipients from suggestions, check your signature in its own collapsible section and drag attachments into your draft
• Ask your mailbox: search permitted older correspondence as well as recent emails, with links to the messages behind an answer. Attachment filenames and file types help you find emails with invoices or other files; the AI does not read attachment contents
• Keep the next step in view: track replies, set reminders and review suggested commitments or contact interest. The AI can also flag signs of spam or fraud for review
• Follow search setup as it runs: the private index builds in the background. A percentage, progress bar and searchable email count update automatically at the top of the mailbox. Search coverage is incomplete until indexing finishes
• Control search and its usage: indexing is enabled for new mailbox connections; existing choices are kept. Indexed text and attachment filenames use your knowledge capacity, while AI requests use your message allowance. You can turn AI search off to remove its index and release the knowledge capacity. Your original emails stay in your mailbox
• Your access settings apply: the workspace uses active email channels, your permitted folders and recipient settings, and your team's mailbox permissions
• Where: Dashboard → Email. Use the settings icon beside the mailbox selector to configure the selected connection

• Every action is now visible in the transcript: a small marker sits above the reply, telling you what happened: "Email contact captured", "Appointment booked", "Website searched", "Live chat requested"
• Open one for the full picture: click a marker to see exactly what the assistant passed on and what came back. Useful when a lead looks wrong or a booking did not arrive
• Your own connectors show up too: tools you connected yourself, whether an API connector or an MCP server, appear under the name you gave them
• Across all channels: web chat, WhatsApp, Messenger, Instagram, email and phone. On the phone it is only recorded when the caller agreed to storing the transcript
• Starts from today: older conversations stay as they were, because this information was never stored before
• Where: Dashboard → Conversations → open a conversation

• Place it exactly where it belongs: choose Left, Center or Right separately for desktop and mobile. The desktop widget opens from its selected side, while a centered button rises naturally from below
• Choose an icon or keep your own image: pick from 12 built-in chat, support and action icons. Seven presets are animated, and the familiar chat bubble with three dots remains the default
• Add attention-grabbing effects with one click: Pulse, Float, Wiggle and Glow are now ready-made options. Pulse creates an expanding ring around the button; reduced-motion preferences are respected automatically
• Better for right-to-left websites: when your page uses <html dir="rtl">, text, messages, menus and controls mirror automatically while your chosen physical button position stays unchanged
• Your existing setup keeps working: uploaded button images and custom CSS remain saved, and older widgets continue to use their current image automatically
• Where: Dashboard → your chatbot → Integration → Widget → Chat button

• Links survive indexing: when a page says "book it here" and links somewhere, your assistant can now pass that link on. Partner, affiliate and booking links keep every parameter exactly as written, so your referral stays intact
• Images in answers: an assistant can show the picture that answers the question, the room, the dish, the product. Only real content images are picked up, not logos, icons or decoration
• Better answers from the same pages: headings, lists and tables are kept during indexing, so a section stays together instead of being cut mid-sentence. Menus and footers no longer end up in the knowledge base, which makes room for the parts that matter
• Documents too: links inside uploaded Word and PDF files no longer get lost on the way in
• What you need to do: re-index a source to pick this up. Existing content keeps working as before until you do, and anything you add from now on is handled the new way automatically

• Point it at a folder once: connect a SharePoint document library, a Google Drive folder, a Nextcloud folder or a directory on your own SFTP server. Everything indexable in it, PDF, Word, Excel, CSV, text, is read into your knowledge base
• It stays current on its own: replace a handbook in that folder and the chatbot knows the new version after the next sync. Delete a file and it disappears from the answers. Nobody has to upload anything again
• Only what changed costs anything: unchanged files are recognised and skipped, so a folder of 500 documents that nobody touched consumes nothing on the next sync. You choose how often it checks: daily, every 3 days, weekly or monthly
• Optional: name the file in answers: switch it on and each answer shows which file it came from, with the path and a link into SharePoint, Drive or Nextcloud. Leave it off for a public website widget, then neither file name nor folder ever reaches the AI
• Where: Dashboard → your chatbot → Knowledge → Add data source, available on Premium and Enterprise. Each dialog links to step-by-step instructions for getting the access details

• Connect your mailbox: link your info@ or support mailbox via IMAP/SMTP (the same credentials as in your mail program) and the assistant answers incoming emails from your knowledge base
• Drafts or direct send: in draft mode replies land in your mailbox as ready-to-send drafts for review; in direct send mode they go out automatically, clearly marked as AI-generated
• Knows the whole history: every reply sees the full prior correspondence with that sender, including emails from before the assistant was connected, and mirrors their tone and form of address
• Safe by design: newsletters, auto-replies and bounces are skipped, unclear requests become drafts instead of automatic sends, and existing mail is never answered retroactively
• Where: Dashboard → your chatbot → Integrations → Email, available from the Standard plan (included in the free trial)

• Popular first questions are answered instantly: when different visitors start a chat with the same question, for example by clicking one of your prompt buttons, the chatbot now reuses its verified answer instead of generating it again
• Does not count against your message quota: repeated answers from this memory are free. Only the first time a question is answered does it count as usual
• Always up to date: as soon as you change your knowledge, settings or prompt buttons, the stored answers are discarded automatically and rebuilt fresh. Answers involving contact forms, appointments or live product data are never reused
• Fully logged: these conversations still appear in your dashboard history and statistics like any other chat
• Where: works automatically for every chatbot, no setup needed

• Mixed-language knowledge now works reliably: many knowledge bases combine languages, an English website with documents in your local language, for example. Until now, a question asked in one language could miss information stored in another
• The chatbot searches in all your languages: it now translates the visitor's question behind the scenes and searches every language of your knowledge base at the same time, then answers in the visitor's language as usual
• Better answers on affordable models: this especially helps the faster, lower-cost AI models find facts they previously overlooked
• Where: works automatically for every chatbot with a multilingual knowledge base, no setup needed

• A clearer new design: the dashboard now has a calmer, more consistent layout. Page titles, menus, buttons and content line up in the same way, so it is easier to understand where you are
• Chatbot settings are easier to find: each chatbot is organized into clear areas such as Overview, Settings, Knowledge, Features and Channels. Advanced options stay available in their own tab without getting in the way
• Test while you work: "Test your chatbot" opens a panel beside your settings. You can keep the preview open while changing colors, text or behavior and see the result without losing your place
• Better on every screen: wide monitors remain easy to read, tables and settings fit better on phones, and Light and Dark mode now have clearer contrast
• Less clutter after setup: installation instructions can be folded away when you no longer need them, while the setup remains available whenever you want to open it again
• Nothing you need to redo: your existing chatbots, settings, website installations and permissions continue to work as before
• Where: the updated design is available automatically throughout the Dashboard

The new WebChatAgent Academy brings all practical guides together in one place. You can learn how to set up your chatbot, connect your own content, design the widget, use forms and integrations, and improve answer quality. Every tutorial uses real product screenshots and shows the finished result.

We will keep adding new guides as WebChatAgent grows. Visit the Tutorial Academy regularly to see what is new.

• Automatic AI labeling: from August 2, 2026 the EU AI Act requires chatbots to disclose they are AI inside the conversation. Every AI reply in your chat widget now carries an "AI Agent" label automatically, in all 13 widget languages
• Honest in both directions: when a human agent takes over the chat, the AI label disappears
• Phone included: the AI phone assistant already introduces itself as an AI at the start of every call
• Where: live in every widget automatically, nothing to configure

• Sticks to your content: chatbots now default to a lower "creativity" setting, so answers stay closer to your knowledge base and invent less, better for support and factual questions
• Still yours to tune: you can raise it per chatbot anytime if you want more varied wording
• Where: Dashboard → your chatbot → Configuration → Temperature

• Block a visitor in one click: Open a conversation's row menu and choose "Block visitor IP", that visitor can no longer use the chat on this chatbot
• Unblock anytime: The new "Blocked IPs" list shows every blocked address with chatbot and date; unblock with a single click
• Where: Dashboard → Conversations (row menu → "Block visitor IP", "Blocked IPs" button next to the filters)

• Filter by date range: Pick any period in the new calendar or jump straight to "This month" / "Last month"
• Export matches your filters: The CSV export now contains exactly what you filtered, chatbot, search and date range, and is much faster for large accounts
• Where: Dashboard → Conversations (calendar button next to the search field)

• Tables scroll instead of squeeze: Wide tables in bot answers now scroll sideways instead of being crushed into the chat bubble, especially on mobile
• Cleaner answer layout: Bot answers avoid unnecessary separator lines and deeply nested lists, so long answers stay readable
• Where: Chat widget, automatic, nothing to configure

• See priority at a glance: Every ticket card now has a colored left bar and a small priority tag, grey Low, blue Medium, amber High, red Urgent, so you can scan a lane in seconds
• New sortable list view: Switch from the board to a table with one click and sort your tickets by any column: priority, due date, assignee, lane and more
• Hide done tickets: Finished tickets in your last lane are hidden by default to keep the board focused; uncheck "Hide done" to show them again. A ticket you just moved to Done stays visible until your next reload
• Where: Dashboard → Tickets (the board/list switch and "Hide done" next to the filters)

• One click, title and description done: When you create a ticket from a live chat, hit "Auto-fill from chat" and the AI writes a fitting title and a clear description straight from the conversation
• Based on the real chat: It reads the actual conversation and summarizes what the visitor needs, including any follow-up points to handle
• Always editable: Everything lands in the form, adjust the title or description before you save
• Where: Dashboard → Live Chats → Create ticket from chat → "Auto-fill from chat" (top of the dialog)

• Readable in dark mode: The ticket description editor now shows your text in a proper light color in dark mode, instead of hard-to-read grey
• Clearer "New ticket" form: Every field now has a label above it, so it's obvious what each dropdown, like Priority, controls
• Where: Dashboard → Tickets → New ticket / ticket details

• A help panel one click away: Click the help icon in the top bar and a panel slides in on the right, with our support assistant ready to answer questions about the platform
• Ask or reach a human: Type your question and the support bot answers right there; need a person? The "Contact us" button opens the support form
• Stays out of your way: The panel pushes the page aside instead of covering it, so you can read and work at the same time
• Where: Dashboard → the help icon in the top bar

• One-click suggested questions for the inline widget too: The "Generate with AI" button that already creates quick-reply buttons for the chat bubble now works for the inline widget as well
• Uses your widget's own context: It reads your widget title and welcome message and proposes fitting questions visitors can tap
• Edit before you save: The AI fills the buttons in for you; tweak anything before saving
• Where: Dashboard → Integration → Inline Widget → Prompt Buttons → "Generate with AI"

• Remove just one page: In the list of indexed URLs you can now delete a single page with the new trash icon, handy for an outdated or irrelevant page you don't want the bot to use
• Delete vs. Exclude: Delete simply removes the page (a later full crawl may pick it up again). Want it gone for good? Use Exclude to keep it off the list permanently
• Frees up space: The page's text no longer counts toward your training data
• Where: Dashboard → Chatbot → Data Sources → Website → Indexed URLs (trash icon per row)

• Various fixes: A round of smaller bug fixes and reliability improvements across the dashboard, widget indexing and account handling

• Style the widget in plain words: Don't know CSS? Just type what you want, like "pulse effect on the chat button", "rounded corners", or "dark header with light text", and the AI writes the styling for you
• Builds on what you have: It reads your current CSS and updates it, so you can refine step by step instead of starting from scratch
• See it instantly, undo anytime: The live preview shows the result right away, not happy? One click on "Undo" restores your previous styling
• One-click examples: A few ready-made prompts get you started in seconds
• Both widgets, Standard and up: Works for the chat bubble and the inline widget
• Where: Dashboard → Integration → Widget / Inline Widget → Advanced → Custom CSS → the "Generate" button

• Make the widget truly yours: When the color and font settings aren't enough, you can now restyle both the chat bubble and the inline widget with your own CSS, rounded corners, custom fonts, a different header, even animations on the chat button
• No !important needed: Your rules automatically win over the built-in styles, so plain CSS like .wca-header { background: #0f172a; } just works
• Stable selectors you can rely on: A documented set of .wca-* classes (header, messages, input, chat bubble, and more) is shown right next to the editor, so your styling keeps working across updates
• Live preview while you type: The preview applies your CSS instantly, check the result before you save
• Available on Standard and up, for both the chat bubble and the inline widget, each with its own CSS field
• Where: Dashboard → Integration → Widget / Inline Widget → Advanced → Custom CSS

• A second way to embed your chatbot: Besides the familiar chat bubble you can now place the chat directly inside your page layout, for example next to a product as an AI product advisor, inside your docs, or in a sidebar
• One line of HTML: Copy the snippet from the dashboard and paste it exactly where the chat should appear, it fills its container and matches your design
• Configure it like the chat bubble: Own settings tab with live preview: title, theme color, font, height, welcome message and image (upload or URL), suggested questions, file upload, sources and automatic translation into the visitor's language
• Built-in chat menu: Visitors can start a new chat, reopen recent chats or end the conversation right inside the widget
• The AI knows which page the visitor is on: Both widgets now send the current page address, visitors can ask "tell me about this product" and the bot prefers knowledge from exactly that page, no extra context needed
• See it live: The inline widget is embedded as a working demo on our AI Product Advisor page
• Where: Dashboard → Integration → Inline Widget

• The Knowledge Tester grew up: It is now the Knowledge Optimizer, one place to test, analyze and improve your chatbot's knowledge, with five tabs: Knowledge Test, Optimize, Role, Model Arena and Audit
• Learns from real conversations: The Optimize tab analyzes past conversations and unanswered questions, groups them by topic and turns them into ready-made content suggestions, review and accept them from a single inbox
• Audit your knowledge base: Scans all your content for duplicates, contradictions and thin content and suggests fixes, contradictions are the most dangerous, because the bot may answer with the wrong version
• A better role, proven: The Role tab proposes an improved assistant role based on failed test questions and proves the improvement with an A/B comparison on the same questions
• Find the best AI model: The Model Arena runs the same test questions against 2–3 AI models in parallel, compares answer quality and speed, and lets you apply the winner to your bot with one click
• Where: Dashboard → Knowledge Optimizer (the former Knowledge Tester menu item)

• Generate the assistant role with one click: In the Custom Role box, hit "Generate with AI" and it reads your data sources and writes a fitting role/persona for your chatbot, so you don't start from a blank page
• Welcome message & speech bubble: Same one-click button for the greeting shown when the chat opens and the short teaser next to the chat button
• Suggested question buttons: Let AI create the quick-reply buttons your visitors click, with sensible questions based on what your business is about
• Always editable: The AI fills the field for you, you can tweak everything before saving. Nothing is published until you save
• In your chatbot's language: Generated texts use the chatbot's Default Language (set under Configuration). Tip: save the Default Language first, the generator reads the saved value
• Plan: Paid plans (Basic and higher)
• Where: Dashboard → Chatbot → Configuration (Custom Role) and Widget Integration (Welcome Message, Speech Bubble, Prompt Buttons) → the "Generate with AI" button

• Reads JavaScript-heavy websites: When a site builds its content with JavaScript (like single-page apps), the crawler now opens it in a real browser to capture the actual content instead of an empty page
• Stopping a crawl is no longer an error: Manually stopping a website indexing run now marks it as "needs reindex" instead of showing a red error
• Status labels in your language: Data source status badges now appear translated
• Where: Dashboard → Chatbot → Data Sources → Website

• Look at any website live: In URL Search, create a group and leave the URL list empty. The chatbot will then read whichever website a visitor sends in the chat, for example to give tips, summarize a page, or answer a question about it
• Only when it makes sense: The bot only opens a website if the visitor's question actually fits what you described in the group. A random pasted link won't trigger it. Write what the bot should do with the website (e.g. "give tips on how to use the chatbot on this site") and that becomes the rule
• Sees more than just text: The bot also picks up the page title, headlines, and the links on the page. So it can follow a link to a sub-page if that helps the answer
• No need to type https://: Visitors can just write "example.com", the bot handles the rest
• Works on tricky pages too: Even when a site renders content with JavaScript or has strict bot protection, the bot still gets enough info to give a useful answer
• Plan: Standard and higher
• Where: Dashboard → Chatbot → Tools → URL Search → add a group, leave the URLs empty

• GDPR-Compliant Auto-Delete: Automatically delete old data after a retention period you define. Set how many days data is kept (default: 30)
• Choose What Gets Deleted: Pick any combination of Conversations, Collected Questions, Feedback, Leads, and Bookings. Anything you don't tick stays untouched
• Daily Cleanup: A background job runs every day and removes data that's past the retention window
• Off by Default: Nothing happens until you enable it. Your existing data stays safe
• Where: Settings → Account → Data Retention

• Welcome Image: Show a large header image above the welcome message in your chat widget
• Custom Avatar & Chat Button: Replace the bot avatar and the closed chat bubble icon with your own image
• Upload or URL: Drag & drop an image directly, or paste a URL
• Where: Dashboard → Chatbot → Widget Integration → Images

• Storage at a Glance: See your storage usage right next to indexed pages, with a color-coded progress bar that turns amber at 80% and red at 100%
• Error Tooltips on Failed Sources: When a data source fails to index, hover over its status to see the exact reason
• Where: Dashboard → Chatbot → Top usage indicator + Data Sources table

• Re-run Same Questions: A new button on any past test result lets you re-run the exact same questions. Perfect for comparing your chatbot's answers before and after a knowledge base change
• More Accurate Evaluation: The AI judge that scores your chatbot's answers now sees the exact knowledge chunks the chatbot used. Result: far fewer false "hallucination" warnings on correct answers
• Real Page URLs in "Find Source": When you click "Find Source", the exact page URL of each match is now shown (e.g. /pricing/enterprise) instead of just the root domain. Click to jump directly to the right page
• No More Timeouts: The detailed reasoning model used for evaluations now has more time, complex test runs no longer cut off mid-evaluation
• Where: Dashboard → Knowledge Tester