Website chatbot

The website chatbot is a public assistant on your own site. Visitors open it from a button on the page and ask questions at any hour. It answers from the knowledge you attach to that chatbot, not from everything your team can see in Centrus chat. In the app it is shown as Custom Chatbot. This guide calls it the website chatbot. The billing card calls it Website Chatbot.

A visitor can ask a question in their own words, or pick from options you set up. The chatbot can take their name and contact details, accept a file such as a photo of a fault, ask for a rating when the chat ends, and pass the conversation to someone in your office. Your team replies from Centrus, or by email if you choose that. The visitor stays on your website the whole time.

It is separate from the chat your colleagues use. A visitor does not sign in, does not see internal conversations, and a reply does not spend your plan's message credits. The chatbot has an allowance of its own, counted for the organisation. You can run more than one, for example one for the main website and another for a careers page, each with its own knowledge and greeting.

This is a standard agent. Centrus builds and runs it. You set it up and point it at your knowledge. You do not build the workflow. See Standard Agents overview.

Every organisation can try the chatbot before putting it on a website. The free allowance is 10 AI message credits a month. Those credits are for the playground, the test chat inside Centrus. Website embedding stays off until you subscribe. One credit is one reply the chatbot writes, whether that reply is in the playground or, once you have subscribed, on your website. A message from a visitor does not use a credit on its own. A reply you type yourself during a handoff does not use one either.

The card is titled Website Chatbot billing and sits at the top of the instances page. Usage this month shows how many credits the organisation has used against the limit, as a count and a percentage. While you are on the free allowance the button is Subscribe at £49.99/month. That turns on embedding and raises the allowance to 1,000 credits a month.

After you have subscribed, the same card says your organisation has full Website Chatbot access, including website embed, and the button changes to Add 1,000 messages at £9.99/month. Each time you add a block, the monthly allowance grows by another 1,000 credits. The blocks stay on the subscription until you remove them from Add-ons.

When the free credits are used up, Free credits used for this month appears. You can still change settings. Playground replies pause until you subscribe or the allowance refreshes. The panel says credits reset on the 1st of the month. The allowance is counted for the organisation over the same monthly period as the rest of your plan. See Plan and usage.

• In the playground, a warning says You have used all 10 free AI message credits for this month while you are on the free allowance, or You have used all Website Chatbot message credits for this month once you have subscribed. Sending another test shows You have reached your Website Chatbot message limit for this month.
• On Share & Embed, the same warning appears, and the embed code stays hidden until embedding is included.
• A visitor on your website does not see those sentences. They see This chatbot is temporarily unavailable. Please contact the business directly.

After a successful subscription the instances page shows Website Chatbot subscription activated. If you leave checkout before paying, it shows Checkout cancelled. In Settings, under Add-ons, Website Chatbot appears on the Active add-ons list once it is subscribed. It is not one of the cards you buy from the General add-ons list. A further row, Website Chatbot messages, appears when you have added message blocks. See Add-ons.

Open Agents from the menu on the left. You need agent access. If Agents is missing, ask an admin. See Edit users and roles. With Agent Builder, standard agents sit at the top of the list, each with a Standard badge and a Use Agent button. Without Agent Builder, each one is a card on the same page. Either way, Use Agent on Custom Chatbot opens it. The Agents page itself is covered in Open Agent Builder.

The next page is the list of instances. The heading uses the agent's description: create and manage website chatbots that answer customer questions using your Centrus knowledge base. Back to Agents returns to the Agents page. New Custom Chatbot starts another instance.

The sort button tooltip is Sort Ascending or Sort Descending. Clear filters appears once you change the search, the filter or the sort. A new instance starts as Personal.

The window title is Create New Custom Chatbot. If you leave the name empty, the page asks you to enter a chatbot name. Cancel closes the window without saving. After a successful create you land back on the list, and Configure Chatbot opens the instance.

Each card shows a badge: Personal, Shared or Organisation. Under the name is the description, or, if you have not written one, Configure this chatbot's knowledge, appearance, embed code and conversation analytics. The date line is Created followed by the date. Configure Chatbot opens the tabs. The copy button duplicates the instance. The bin deletes it.

• A duplicate is named with (Copy) on the end of the original name, keeps the description, and starts as Personal. The knowledge, actions, appearance and embed settings are not copied. Set those up on the new instance. The confirmation is Chatbot instance duplicated successfully.
• Only the person who created the instance can delete it. Confirm Deletion asks Are you sure you want to delete the name, and warns This action cannot be undone. Delete removes it and shows Chatbot instance deleted successfully.
• Anyone else who clicks the bin sees Permission Required: they do not have permission to delete it, and only the creator can. The window names the creator so they know who to ask.

If nothing matches the search, the page says No matching instances found and offers Clear Filters. If the organisation has no instances yet, it says No instances created yet and Create your first chatbot instance to start building a website assistant, with Create Custom Chatbot.

Configure Chatbot opens the instance. The title is Custom Chatbot, then the instance name. Under it, if you have written a description, that description is shown. Back to Instances returns to the list. Along the top are seven tabs: Settings, Knowledge, Actions, Customisation, Playground, Share & Embed and Analytics. A link that ends with the actions tab opens Actions directly. The cards on Settings start collapsed. Click a heading to open it.

Chatbot Details holds two fields. Chatbot Name is the internal name, with the note This is the internal name shown to your team in Centrus. It is not the title visitors see. That title is set under Customisation. Chatbot Description is a note for colleagues. The placeholder is Describe what this chatbot is for, for example: Main website chatbot for customer enquiries. The help line says the description is for internal users, and that public behaviour is configured on the Knowledge and Customisation tabs. Save Settings stores both. The confirmation is Chatbot settings saved successfully.

Sharing & Permissions decides who else can open this instance. The switches are only available to the creator. Everyone else sees them disabled.

• Visible to entire organisation. When this is on, everyone in the organisation can open the instance. Turning it on for the first time also sets members to view only, until you allow editing. The confirmation is Chatbot is now visible to your entire organisation. Turning it off limits the instance to you and anyone you have picked, and the confirmation is Chatbot is now only visible to you and selected team members.
• Organisation members can edit. This appears only while the instance is visible to the organisation. On means members can change it. Off means they can view it and only the creator can edit it, which is what the help line says. The confirmations are Organisation members can now edit this chatbot and Organisation members can now only view this chatbot. A line under the switch reports Organisation permission: Can edit or Organisation permission: Read-only.
• Field Permissions, beside that switch, lets the creator hide Chatbot Name or Chatbot Description from members who can otherwise edit. Those are the only two fields it covers.
• Share with selected team members. This appears when the instance is not visible to the whole organisation. Turn it on and pick people. The creator is not in the list. Each change to the list is saved straight away. For each person you can allow editing or leave them as view only. The confirmation for a permission change is User permission updated.

If you cannot edit, Save Settings stays disabled and a save attempt shows You do not have permission to edit this chatbot. The same sentence appears on the Knowledge and Customisation tabs.

The Knowledge tab is where the chatbot learns what to say and which documents it may use. There are three cards, each collapsed until you open it: Personality & Instructions, Knowledge Base Tags and Structured Information. Nothing on this tab is shown to visitors as a settings screen. It shapes the answers. Save Knowledge Settings at the bottom stores the whole tab. The confirmation is Knowledge settings saved successfully.

Personality and Instructions is a single box. The placeholder reads Example: You are a helpful, professional website assistant. Keep answers concise, friendly and focused on helping customers understand our services. The help line says this controls tone and behaviour, and that the chatbot will still use the selected knowledge tags and information blocks as its source material.

Name the company, the tone, and the next step. For Meridian FM: you are the website assistant for Meridian FM. Answer in short sentences. If someone wants a visit, offer to take their name, site and phone number. Do not quote prices. Avoid a vague line such as be helpful and estimate prices. That fights the guardrails.

Guardrails starts on. The help line says it keeps answers on approved business information. Guardrail Instructions is shown while the switch is on, and kept when you turn it off. The starting text says to use only the instructions, the configured information and the knowledge base files, not to invent prices, policies, availability, opening hours or legal or financial advice, and to say so and suggest contacting the team when it is not sure. Make it specific: only Meridian FM services, covered sites, and the hours in Structured Information. Do not confirm a booking or give a price. Be accurate does not say what to refuse.

Knowledge Base Tags chooses the documents. File Search is ticked and cannot be unticked, with a Knowledge Base badge. Under Knowledge Base Access you pick one or more tags. The placeholder is Select one or more tags, or No tags available when the list is empty. The help line is Select the tags containing the documents this website chatbot should use. If you have no tags you can use, it says No knowledge base tags are available to your user.

The chatbot searches only the files on the tags you select. An admin sees every tag. Anyone else sees the tags on their own user. If your user has no tags, the list shows the organisation's tags, so check it before you save. See Control who sees what with tags, Upload files and Train from a website. Use tags for documents already in Train. Use the next card for facts you want written out, such as this week's opening hours.

Structured Information is text you type, used with the documents. Add Templates inserts General Business Information, Opening Hours and Services, with the fields already named. Add Section inserts New Information Section. An empty card says No information sections yet. Each section has Section Title (placeholder e.g. Opening Hours) and Section Description (placeholder Optional internal note for this section). Add Field adds New Field, with Field Name (e.g. Monday) and Field Description (e.g. 09:00 to 17:00). The bin removes a field. Delete Section removes the section. Put the fact in Field Description. The section description is only a note for your team. For Meridian FM, Monday to Friday might be 08:00 to 17:30, and Call-out cover the evenings and weekends for Northgate Housing at Riverside House.

The Actions tab holds six cards: Integrations, Guided Conversation Tree, Human Handoff, Attachments, Conversation Feedback and Lead Capture. They start collapsed. The first one visitors notice, if you turn it on, is the guided conversation: a set of buttons under the opening message, so someone can choose Book a visit or Ask about cover without typing.

The card description says Enabled or Disabled, then structured option blocks for embedded and playground chats. Enable Guided Conversation Tree shows the buttons before visitors type freely. First Assistant Message starts as Hello! How can I help you today? The help line says it overrides the standard welcome message while the tree is enabled. The welcome message on the Customisation tab is used again if you switch the tree off.

Root Option Block is the first set of buttons. Add Block creates another. Each block has an Internal Block ID, an Assistant Message (placeholder What should the assistant say before showing these options?) and options. Add Option adds New option. An empty block says No options in this block yet.

• Show another option block. Next Option Block picks which set opens next. Use this for a short path, such as Book a visit, then Planned maintenance or Call-out.
• Send a prompt to AI. Prompt Sent To AI is the question sent on the visitor's behalf. AI Response Instructions is optional and tells the chatbot how to answer that one button.
• Show a fixed message. Assistant Message is shown as written. The chatbot does not write a new answer for that click.
• Open a URL. URL is the address, with Open in new tab ticked by default.
• Show message and open URL. Needs both the message and the address.
• Open lead capture form. Shows the lead form. Lead capture has to be switched on, or the check below warns that the button opens lead capture but lead capture is disabled.

Amber text headed Guided conversation tree checks lists problems before you save: a root that does not point at a block, a block with no message and no options, a button that links to a missing block, a prompt or message or address that is empty. Save Guided Conversation Tree stores the tree together with the other action settings, and shows Action settings saved. Keep to 30 blocks, and to 10 options in any one block. A tree that breaks those limits, or that still has a broken link, is not kept. After you save, open the card again and confirm the blocks are still there.

Visitors can move back up the path. Go back appears from the second step, and is hidden on the first. You can hide it, and restyle the buttons, under Guided Options on the Customisation tab. That section says to enable Guided Conversation in Actions before those controls apply in the live widget.

Lead Capture collects a name, an email and anything else you ask for, then stores the submission on this card. The heading counts new leads and the total, for example 1 new lead and 4 captured leads in total. Inside, Lead Settings is where you switch it on, and Captured Leads is the list.

Enable Lead Capture lets the chatbot show a form to visitors with buying or enquiry intent. Lead Capture Instruction tells it when. The placeholder is Tell the AI when it should request visitor lead details. The starting text asks for details when the visitor shows commercial interest, asks for a quote, requests a callback, asks about pricing, or wants to discuss a service, and tells it not to ask during an unresolved support issue or while a handoff is active. For Meridian FM you might add: also ask when someone wants a visit booked at a Northgate Housing site.

Trigger Mode has three choices. AI detected lead intent shows the form when the chatbot decides the conversation has reached that point. Manual form button only never shows it on its own. The visitor uses a Leave your details button above the message box, which appears once the chat has started. AI detected and manual button does both. Form Title starts as Request a Callback. Intro Message starts as Leave your details and the team will get back to you. Success Message starts as Thanks, your details have been sent to the team. The visitor sees those three sentences on the form and after they submit. The button on the form is Submit Details, and it shows Submitting... while it sends.

Consent Checkbox is on to start with. Consent Text starts as I agree to be contacted about my enquiry. The visitor must tick it before the form will send, when the switch is on. Hide the box by turning the switch off. The consent text field is hidden with it.

Form fields is the form itself. The count reads, for example, 5 of 20. A new chatbot starts with Name, Email, Phone, Company and Message. Name, Email and Message are required. Phone and Company are optional. The help line says visitors see exactly what you configure here. You can rename a field, drag it to a new position, mark Required, or remove it. Reset defaults restores those five. Add common fields adds a ready-made set. Add field adds a blank one, up to 20 fields.

• Field types are Text, Long text, Email, Phone, Postcode, Dropdown and Attachment.
• A Dropdown needs its choices. You can add up to 20.
• An Attachment field accepts JPEG, PNG, WebP and PDF to start with, and you can also allow Word documents and MP4, MOV or WebM video. Up to 5 files, each up to 3 MB, so the file can be included on the alert email.
• An empty label is highlighted. The page will not save until every field has a name.

Lead email notifications uses a mailbox from Integrations. Notify team on submission emails Team notification recipients (placeholder sales@example.com, enquiries@example.com, commas between addresses), including files up to 3 MB each. Send customer confirmation email writes to the address on the form. Without an email field, an amber note tells you to add one and mark it required. Mailbox provider is Microsoft Outlook or Google Gmail. Customer confirmation subject starts as We received your enquiry. The message can include {{name}}, {{email}}, {{phone}}, {{company}}, {{message}} and {{submission_summary}}. Save Lead Capture stores the card.

Captured Leads says No captured leads yet when empty. A row shows a name, or Unnamed lead, the other details, NEW, REVIEWED or CLOSED, and the message or No lead message provided. The status menu is New, Reviewed or Closed. View Conversation opens a read-only transcript titled Feedback Conversation. Ten leads show at a time, then Previous, Next and a page line.

Human Handoff lets the chatbot stop answering and wait for a person. The card counts open conversations, for example 1 open conversation and settings for routing visitors to your team. Handoff Settings is the setup. Handed Off Conversations is the queue.

Enable Human Handoff allows the chatbot to switch an embedded chat into a live conversation with your team. Handoff Channel is Centrus App or Email. The help line says to choose whether your team handles handoffs in Centrus, or via email alerts with reply-by-email back into the chat. Centrus App is the starting choice. Your team opens the conversation on this tab and types a reply. Email sends an alert from a mailbox you have connected, and a reply from one of the alert addresses is delivered back into the visitor's chat.

Email adds Email Provider (Microsoft Outlook or Google Gmail; connect it under Integrations first), Alert Recipients (placeholder team@example.com, manager@example.com; replies from these addresses come back into the chat), Inactivity Auto-Close (5 minutes, 15 minutes, 30 minutes, 1 hour, 2 hours, 4 hours; starts at 30 minutes) and Handoff Email Prompt (placeholder Describe how the AI should format the handoff alert email for your team). The help icon says the alert can include the transcript, the guided path, lead details and the handoff context.

Trigger Description says when to hand off. The placeholder is Describe when the AI should hand off to a human. The starting text hands off when the visitor asks to speak to a person, needs something outside the chatbot's knowledge, is frustrated, or has a sales or support request that needs a team member. Trigger Keywords are extra hints, comma-separated. The placeholder is human, support, call me, speak to someone. The starting list is human, person, agent, team, support, call me, contact me, speak to someone. The help line says they guide the chatbot alongside the trigger description. Visitor Handoff Message is what the visitor sees when the chat is passed over. It starts as I have handed off the conversation. The staff member will reply as soon as possible. Rewrite it in your own voice. Save Human Handoff stores the card.

The visitor sees the handoff message, then your replies in the same window, whether you type them in Centrus or reply to the alert email. If an email handoff sits idle past Inactivity Auto-Close, their next message is This conversation was closed due to inactivity. Please start a new chat if you still need help. and the chat resets. A closed handoff will not take another message until they start again.

Handed Off Conversations says how many open conversations need a reply. An empty queue says No handed-off conversations yet. Each row has a badge, OPEN or CLOSED, the date and time, the last message or No message content yet, and the website, or Embedded chatbot if no address was recorded. Open Conversation opens Human Handoff Conversation. The subtitle says it polls every 5 seconds while open for new website visitor messages.

The window shows Status, Last message order, Mark Closed or Reopen, and messages labelled End User, AI Assistant, Human Agent or System. Files sit above the transcript. Reply as Human Agent uses Type your reply... while open, and Reopen the conversation to reply while closed. Send delivers the reply. A failure shows Failed to send message.

Attachments adds a paperclip to the message box, in the playground and on the website. The card says Enabled or Disabled. It starts enabled. Enable attachments shows the paperclip. Max files per message starts at 5 and accepts 1 to 10. Max file size (MB) starts at 10 and accepts 1 to 25.

Allowed file types is a row of tick boxes. Ticked to start with: JPEG images, PNG images, WebP images, PDF documents, Word documents (.doc) and Word documents (.docx). Also available, unticked: MP4 video, MOV video and WebM video. Video types have a help icon: The AI will not be able to see the contents of the video. You must leave at least one type ticked. Clearing them all shows Select at least one allowed chat attachment type. Save Attachments stores the card.

Files are kept for 30 days and then removed. There is no control on this card to change that period. In a handed-off conversation and in Analytics, the files for that chat are listed above the transcript so your team can open them while they are still kept. If attachments are switched off, a visitor who already has the widget open cannot upload, and the chatbot refuses a file with Attachments are disabled for this chatbot.

Conversation Feedback asks the visitor for a rating out of five when a chat winds down. The card description counts the ratings and, once there is at least one, the average, for example 6 ratings with 4.5/5 average feedback. With none yet, it says the ratings are collected from embedded conversations. Feedback Settings is the setup. Collected Feedback is the list.

Enable Conversation Feedback lets the chatbot request a 5-star rating when it detects the chat has naturally ended. Feedback Instruction is when to ask. The placeholder is Tell the AI when it should request feedback. The help line says this works like a guardrails prompt and guides when the feedback screen should appear. The starting text says to ask only when the visitor seems satisfied and the conversation has ended, with signals such as thanks, that is all, bye, perfect, no more questions, or a clear resolution, and not to ask during an unresolved issue, in the middle of support, or while a handoff is active.

Trigger Mode is AI detected ending, Inactivity only, or AI detected ending and inactivity. Comment Box is on to start with, and lets the visitor add optional written feedback. Feedback Prompt starts as How was this chat? Thank You Message starts as Thanks for your feedback. The visitor sees the prompt, the stars, the comment box if you left it on, and then the thank you message. Save Conversation Feedback stores the card.

Collected Feedback counts public submissions. An empty list says No feedback collected yet. Each row shows the stars, the date and time, the comment or No written comment, and the website or Embedded chatbot. View Conversation opens the same read-only transcript used for leads. The list shows 10 ratings at a time, with Previous, Next and a page line once there are more.

Integrations, the first card on Actions, connects a mailbox for this instance. It is not a connection to a job system, and it does not give the chatbot new tools. The description says to connect Outlook or Gmail mailboxes for this chatbot instance, used for email handoff and lead notifications. The help line says to connect one or both, and that human handoff and lead capture each choose which connected provider sends alert emails.

There are two rows, Outlook and Gmail. Each shows Checking Outlook connection... or the Gmail equivalent while it looks. Not connected means you still need to connect it for email handoff alerts and lead notifications. Connect starts the sign-in. After it succeeds the page shows Outlook connected successfully or Gmail connected successfully, with the mailbox address when one comes back, and returns you to the Actions tab. A connected row says Connected to and the address, or Connected to mailbox if the address is missing. Disconnect removes it and shows Outlook disconnected or Gmail disconnected.

Connect the mailbox before you rely on Email as the handoff channel, or on lead email notifications. The provider you pick on those cards has to be one of these two, and it has to be connected. A failed connect or disconnect names the mailbox in the message, for example Failed to connect Outlook.

Customisation is the look of the button and the chat window, separately for desktop and for mobile. The card is Chatbot Customisation, and the description says it controls desktop and mobile widget appearance, layout, typography, animations and message styling. Desktop and Mobile switch which set of values you are editing, and which preview you see. Reset puts every value back to the defaults and shows Customisation reset to default values. Save to apply. Nothing changes for visitors until you click Save Customisation, which shows Customisation settings saved successfully. While it saves, the button reads Saving....

The preview stays on the right and matches the device you have selected. A new chatbot uses a navy header, white text, the title Website Assistant, the welcome Hello! How can I help you today?, the placeholder Ask me anything..., and the line Powered by Centrus AI under the input. The button sits at the bottom right. Change Window Title to your own name, such as Meridian FM, so visitors know who they are talking to.

Where a control offers Custom for an icon, an upload row appears: No custom image uploaded, then Upload, or Replace once a file is there. The help line is PNG, JPG or WebP. Recommended square image under 2 MB. Any other file shows Please upload a PNG, JPG or WebP image. A file of 2 MB or more shows Please upload an image smaller than 2 MB. The same rule applies to the header icon, the bot avatar and the user avatar.

• Chat Button: Button Colour, Button Hover Colour, Button Size (Small, Medium, Large), Button Position (Bottom Right, Bottom Left), Button Icon (Custom, Bot, MessageCircle, Headphones, Users, Crown, Shield, User), Icon Size (12 to 72), Border Radius (0 to 60), Bottom Offset, Side Offset (0 to 120), Shadow (None, Light, Medium, Heavy), Button Animation (None, Bounce, Pulse, Shake, Glow), Button Animation Interval (ms) (1,000 to 30,000).
• Chat Window & Layout: Window Title, Welcome Message, Input Placeholder Text, Header Icon, Header Icon Size (12 to 48), Window Width (260 to 560), Window Height (360 to 820), Window Border Radius, Window Shadow, Header Height (36 to 90), Chat Bottom Offset, Chat Side Offset, Welcome Message Delay (ms) (0 to 10,000), Entrance Animation (Slide Up, Slide Down, Fade In, Scale Up, None). Mobile Fullscreen is only on Mobile.
• Colours & Branding: Header Background, Header Text, AI Message Background, AI Message Text, User Message Background, User Message Text, Input Background, Show Powered By (on hides or shows Powered by Centrus AI).
• Guided Options: Show Guided Options, Show Go Back Button (hidden on the first step), Option Background, Option Text, Option Border, Option Hover, Option Border Radius (0 to 32).
• Typography: Font Family (System Sans, Inter, Roboto, Poppins, Serif), Header Text Size (10 to 32), Header Font Weight, Message Font Weight (Light, Normal, Medium, Bold), Message Text Size (8 to 24), Placeholder Text Size (8 to 20), Branding Text Size (8 to 16).
• Messages & Avatars: Message Border Radius, Message Max Width (%) (50 to 100), Message Shadows, Show User Avatar, Show Bot Avatar, Avatar Style (Circle, Rounded, Square), Bot Avatar Icon, User Avatar Icon, Message Animation (Fade In, Slide In, Scale In, None), Thinking Colour.
• Input & Send Button: Input Border Radius, Input Size, Placeholder Text Colour, Input Text Colour, Send Button Colour, Send Button Hover Colour, Send Button Size, Send Button Icon (Send, ArrowRight, ChevronRight, Play, MessageCircle), Send Button Icon Colour, Send Button Border Radius.
• Advanced Behaviour: Loading Animation (Spinner, Dots, Bars), Loading Spinner Colour, Loading Text Colour, Loading Background Colour. Changing these plays the loading state in the preview. Show Timestamps starts off. Timestamp Format is 12 hour, 24 hour or Relative.

The Playground tab runs the saved chatbot without putting it on a website. The title is Playground, with an Internal test badge. The description says to test the chatbot using its saved knowledge, guardrails and customisation before embedding it on a website. Desktop and Mobile change the frame. New Session starts a fresh conversation. Clear empties the messages on screen. Each reply the chatbot writes here uses one credit, the same as a reply on the website. When the allowance is used up, the warning at the top of the tab blocks further replies.

If you have not selected any tags and have not added structured information, a warning says No knowledge has been configured yet, and tells you to add tags or structured information on the Knowledge tab before relying on test responses. You can still send a message. The answers will have little to go on.

Live Chatbot Test says to click the widget to test the saved website chatbot experience. On desktop the widget sits on a plain background. On mobile it sits inside a phone frame. It uses the colours, title and welcome you saved. Quick Tests offers six prompts: What services do you offer?, What are your opening hours?, How can I contact your team?, Where are you based?, What do you do? and Can I get a quote? Clicking one sends it. Runtime Inputs lists how many knowledge tags and information sections are saved, whether guardrails are Enabled or Disabled, a short conversation reference or New, and the handoff status.

Playground conversations are kept. On Analytics they are marked Playground, and the website line reads Internal playground when no site was involved. Use the playground to check opening hours, a guided path and the lead form before you publish the embed code. Save knowledge and customisation first. The playground reads what is saved, not unsaved edits.

Share & Embed is where the chatbot becomes public. Until the subscription includes embedding, the card says Upgrade to publish this chatbot on your website, the code is hidden, and a note says website embed is available on the paid Website Chatbot plan at £49.99 a month, which includes 1,000 AI message credits per month, with Upgrade to publish on your website. The switch stays disabled.

<script src="https://app.centrus.ai/embed/custom-chatbot.js" data-chatbot-id="mfm8k2website"></script>

Once embedding is included, Website Embed Code says to copy the one-line install code. The switch reads Enabled or Disabled. It starts enabled. Turning it off and saving shows an amber note: Website embedding is disabled. Existing embed codes will stop loading until this is enabled again and saved. The help line says to paste the recommended one-line script before the closing body tag, or into your website builder's custom code area.

Simple Embed (Recommended) is one script tag, labelled HTML, with Copy. Success is Embed code copied. A blocked copy is Failed to copy embed code. The script loads from your Centrus app and carries this instance's public id. With a made up id, the source is https://app.centrus.ai/embed/custom-chatbot.js and the chatbot id is mfm8k2website. Advanced Embed adds position bottom-right and open false, so the window starts closed. The button position from Customisation still applies.

Open Preview opens a preview of this instance in a new tab. It stays disabled until a public id exists and the switch is on. If you click it while embedding is off, the page says Enable website embedding and save before opening preview. If the id is not ready, it says Embed public ID is not available yet. Save Embed Settings stores the switch and the domain list, and shows Share & Embed settings saved.

Allowed Domains uses the placeholder **example.com or *.example.com. Enter or Add Domain** adds it. A full address is trimmed to the host, so a contact page on www.meridianfm.co.uk becomes meridianfm.co.uk. *.meridianfm.co.uk also allows subdomains. The x removes a pill. Save afterwards. An empty list says No domain restrictions are enabled. Any website with the embed code can load this chatbot. Add meridianfm.co.uk before you publish. A site that is not listed is refused with This website is not allowed to use this chatbot. Centrus checks the address of the page the chatbot opens on, as the visitor's browser reports it. If your own site is refused, check that the domain in the list matches the one in the address bar, and ask whoever runs the website whether it stops browsers sending the page address to other sites; with that turned off, a listed site cannot be confirmed and is refused.

While the embed is loading, a failure shows Chatbot embed is not available if the instance is switched off, the id is wrong, or the code has been removed. A visitor who sends messages unusually quickly sees Too many messages. Please try again shortly. The same kind of limit applies to uploads, feedback and lead forms. A single message can be up to 2,000 characters. Longer text is refused with Message is too long.

On meridianfm.co.uk the visitor sees a round button at the corner you chose, bottom right unless you moved it. Clicking it opens the window. The header is the window title, Meridian FM in this example. If the guided conversation is on, the first assistant message appears with the buttons underneath, and the welcome message from Customisation is not used. If the guided conversation is off, the welcome message appears on its own, after the delay you set, which is immediate by default.

The visitor can type in the box, whose placeholder is Ask me anything... unless you changed it, and send. If attachments are on, a paperclip sits in the box. If lead capture is set to the manual button or to both, Leave your details appears above the input once the chat has started. Choosing a guided option either opens the next buttons, shows a fixed message, opens a page, sends a question for the chatbot to answer, or opens the lead form. Go back returns to the previous set of buttons.

A lead form uses your title, intro, fields and consent text, then Submit Details, then the success message. A handoff shows the visitor handoff message and then waits. Replies from your team appear in the same window. A finished chat can show the feedback prompt, five stars, the comment box, and the thank you message. If the monthly allowance is used up, the visitor sees This chatbot is temporarily unavailable. Please contact the business directly. They are not told about credits or prices.

On a narrow screen the window uses the mobile values from Customisation: its own title, welcome, colours, size and position. Mobile Fullscreen makes it cover the page. Test both with Desktop and Mobile in the playground, and with Open Preview, before you treat the snippet as live. The preview and the playground do not replace a check on the real page, because allowed domains and the snippet only apply there.

Analytics records playground tests and website chats. The counters are Today's Messages, Today's Conversations, Total Messages and Total Conversations. Today means conversations updated today, and today's messages adds up the full message count of those chats, including earlier messages. The totals follow the filters. The list opens on Last 30 days, so the totals start there, not over the life of the chatbot. Search and source change the counters too.

All Conversations is collapsed until you open it. The description counts conversations across embed and playground sessions. The filters are Search keywords (placeholder Search messages, domains or URLs), Date modified (All time, Today, Last 7 days, Last 30 days, Custom), Source (All sources, Embed, Playground), Status (All statuses, Active, Completed, Errored, Archived), Sort and Handoff. Custom adds Start date and End date. Apply runs the search. Reset clears it back to the defaults, including Last 30 days and newest modified.

Sort is Newest modified, Oldest modified, Newest created, Oldest created, Most messages or Longest response time. Handoff is All handoffs, No handoff, Requested, Open or Closed. The list says Loading conversations... while it fetches, and No conversations match these filters yet when nothing does. Each row shows Embed or Playground, the status, Handoff: and the status when there was one, Modified and the time, the first message or Conversation started, the last message or No message preview available, the message count, Average response time, and the domain, the page, or Internal playground. View Conversation opens the detail. The list is paged, and you can change how many rows appear.

Conversation Details says it is a read-only transcript and session metadata. It shows Source, Status, Handoff (or none), Messages, Created, Modified, and the domain or No website domain recorded. Roles in the transcript are End User, AI Assistant, Human Agent and System. You cannot reply from this window. Replies to an open handoff are on the Actions tab, under Handed Off Conversations.

These are the messages the product shows when something needs a decision or has failed. The visitor-facing lines are included because your team will be asked about them.

Opening Agents needs agent access. Inside an instance, what you can change depends on who created it and how it is shared. Being an admin does not, by itself, let you edit or delete someone else's chatbot. Admins do see every knowledge tag when they choose what the chatbot may read.

Credits are counted for the organisation, not for one person. A colleague testing in the playground spends the same allowance as a visitor on the website. If several people are testing on the same day, check Usage this month before you assume the website chatbot has stopped on its own.