Understand suggested questions
Choose the questions visitors see and regenerate suggestions after updating knowledge.
Updated 2026-10-10
A white-label chat bubble for your site
Send and receive through your own system
Bring customer email into one inbox
Connect WhatsApp Business to serve customers
Connect Facebook Pages and handle messages
Receive and reply to Instagram DMs
Connect a LINE official account to serve customers
Connect a Telegram bot for customer messages
Connect a Zalo Official Account
Receive and reply to comments
Connect official enterprise customer service accounts
Authorize securely with Facebook
No App Secret, verify token, or Page token to copy. YundaDesk lists manageable Pages and checks the connection automatically.
Authorize myself
The current staff member manages the Facebook Page.
Connect with FacebookSelect a Facebook Page
Complete authorization in the new tab. It closes automatically and returns the result to this workspace.
YundaDesk subscribes messages, postbacks, delivery, and read events, then runs a live connection check.
Finish connectionChoose your email provider
Microsoft OutlookConnect an Outlook or Microsoft 365 mailboxChannel maintenanceMail settings are filled in from your address. You can edit them.
No preset matched. You can enter the mail settings manually in the next step.
Advanced settings
imap.yourbrand.com · 993smtp.yourbrand.com · 465Save disabled first, test IMAP and SMTP, and enable the channel only after both pass.
Connect & enableSelect customer service accounts
Each selected account becomes an independent channel that is enabled immediately, with AI replies off by default.
DEMO_REDACTED_IDENTITY_KEYWhen off, customers wait for a team member.
View the customer identity signing key in channel settings. Your website backend signs an HS256 JWT for the current signed-in user. Keep the key on your backend only.
Include user_id, widget_key, iat, and exp; set aud to yundadesk-widget-identity. Use Unix seconds for timestamps. A 5-minute token lifetime is recommended.
Run this code after website login. Replace /support/identity with your website endpoint, which returns { token }.
const loginSupport = () => YundaDesk.login({
getToken: () => fetch('/support/identity', { credentials: 'same-origin' })
.then(response => {
if (!response.ok) throw new Error('Identity request failed');
return response.json();
})
.then(({ token }) => token),
});
if (window.YundaDesk?.login) loginSupport();
else window.addEventListener('yundadesk:ready', loginSupport, { once: true });Call YundaDesk.logout() before website logout or account switching to clear the customer identity in the Widget.
Optional fields: name, email, phone, avatar, locale. custom_attributes accepts enabled customer custom fields only; use their field keys.
<script src="https://cdn.yundadesk.com/widget.js" data-yd-key="yd_widget_main" async></script>- Widget Key
- yd_widget_main
- Script URL
- https://cdn.yundadesk.com/widget.js
- Direct link
- https://cdn.yundadesk.com/?widget_key=yd_widget_main
Telegram bot setup guide
Three steps: ① find @BotFather to create a bot and get the token → ② paste the token on the left → ③ save and enable; the system receives new messages automatically.
1. Create a Telegram bot
- In Telegram, search for @BotFather and open the chat.
- Send /newbot to create a new bot.
- Follow the prompts to set a display name and username (ending in _bot).
- Copy the bot token returned by BotFather.
2. Enter the token on the left and enable
- Paste the token into “Bot Token” on the left.
- Click “Save” at the top, then “Enable”.
- Click “Test connection” to verify the bot token works.
3. Test the bot
- Search for the bot username in Telegram and start a chat.
- Send any message and AI / an agent will reply automatically.
- See Telegram Bot API docs.
Inbound webhook URL
Copy this full URL into the webhook settings of the third-party console so customer messages reach YundaDesk.
https://api.example.invalid/hooks/channel/line/demo-onlyCopyLINE Official Account setup guide
1. Enable the Messaging API
- Sign in to LINE Official Account Manager, select the official account to connect, and enable the Messaging API under Settings > Messaging API.
- Copy the Channel ID and Channel Secret from the same page.
2. Save and enable in YundaDesk
- Enter the Channel ID and Channel Secret on the left and save. YundaDesk verifies them with LINE and then shows the matching LINE Official Account.
- Click Enable so the channel starts receiving messages.
3. Set up the webhook and test
- Copy the inbound webhook URL above, paste it into Webhook URL under Settings > Messaging API in LINE Official Account Manager, and save.
- Turn on Webhook under Settings > Response settings. We also recommend turning off auto-response and greeting messages so customers don't get duplicate replies.
- Back in YundaDesk, click Test connection. Once it passes, send a message to the official account from LINE and check that it appears in Chat.
The channel credential has expired. Please reauthorize and try again.
Connection failedBot receiving conflict
Bot webhook is active
Telegram bot setup guide
Three steps: ① find @BotFather to create a bot and get the token → ② paste the token on the left → ③ save and enable; the system receives new messages automatically.
1. Create a Telegram bot
- In Telegram, search for @BotFather and open the chat.
- Send /newbot to create a new bot.
- Follow the prompts to set a display name and username (ending in _bot).
- Copy the bot token returned by BotFather.
2. Enter the token on the left and enable
- Paste the token into “Bot Token” on the left.
- Click “Save” at the top, then “Enable”.
- Click “Test connection” to verify the bot token works.
3. Test the bot
Callbacks go through YundaDesk fixed secure egress. If your receiver restricts sources, add the egress IPs to its allowlist.
Callbacks go through YundaDesk fixed secure egress. Allowlist these IPs: 203.0.113.10, 203.0.113.11
View allowlist IPsCustom API setup guide
1. Create a Custom API channel
- Enter the channel name and “your system callback URL” on the left.
- After saving, the system generates a connection key and signing key to confirm message ownership and request authenticity.
- (Optional) Add an IP allowlist to restrict sources.
2. Forward customer messages to YundaDesk
- POST messages as JSON to the webhook URL above.
- Include the connection key, timestamp, signature, and idempotency key in the headers.
- A 200 response means received; other statuses are treated as failures and retried.
3. Receive reply pushes
- Once enabled, YundaDesk pushes agent / AI replies to the “your system callback URL” you entered on the left.
- Deduplicate by external message ID for idempotency.
| Event | Time | Status |
|---|---|---|
| message.created | 2026/07/28 14:20 | Delivered |
| message.created | 2026/07/28 14:06 | Timeout · retrying |
| conversation.assigned | 2026/07/28 13:41 | Signature rejected |
Publish a public help center so customers can find answers on their own and cut repeat questions.
Install one YundaDesk plugin to add live chat, AI support, and a secure Widget binding to a WordPress site.
Use the same YundaDesk WordPress plugin; after WooCommerce is detected and authorized read-only, sync store, customer, order, product, and current cart context.
Connect SHOPLINE to sync orders, customers, and products, and enable YundaDesk storefront chat in the theme when needed.
Connect a BigCommerce store, sync customer order context, and show recent orders in the workspace sidebar.
Can’t find what you need?
Store actions awaiting approval
Review the store-generated preview before approval. Approval triggers one controlled commit; changed inputs require a new preview.
Create recovery discount · DEMO-SAVE10Live store preview · 10% · Demo Yundastore
The store has not returned a definitive result. Do not submit it again; this control only queries the result and never repeats the action.
Check resultRequest
Inputs
Outputs
Usage rules
☑ Require verified customer identity
Only customers verified in the current conversation can use this function. Otherwise, execution pauses.
Request
Enable only if this POST endpoint queries data without changing it.
Inputs
Parameter settings
Sent as a JSON body field. Enter valid JSON and describe its structure in the parameter description.
Request
Inputs
Usage rules
Result verification (optional)Use the API response by default. Add a read-only check if further verification is needed.
View tests and Agent calls for this function.
Select tools for future use. Saving does not grant Agent execution permission.
Credentials are stored securely and never displayed.
Connectors
Choose the functions this Agent can use. Calls follow function settings and business rules in skills.
Search connections or functions
☑ lookup_order
☐ lookup_order Select the output fields available to the Agent first.
☐ lookup_refund Configuration changed. Select again to authorize.
| Customer | Channel / source | Location / language | Tags | Last interaction | Added | |
|---|---|---|---|---|---|---|
| MMaria Garciamaria@example.com | Web Widget | 🇪🇸 SpainSpanish | 05/22/2026 | 05/18/2026 | ||
| LLukas Schneider+49 30 123456 | Telegram | 🇩🇪 GermanyGerman | 05/20/2026 | 05/16/2026 | ||
| AAna Silva | 🇧🇷 BrazilPortuguese | 05/22/2026 | 05/16/2026 |
| Agent | Role | Reception channels | Readiness | Status |
|---|---|---|---|---|
| AAlexManaged AI | General reception | Website · WhatsApp | Ready | Running |
| NNovaManaged AI | General reception | Messenger | Ready | Running |
| SSamExternal AI | General reception | — | Setup incomplete | Draft |
Choose which tools and actions this Agent may call. These switches never replace runtime authorization, approval, or audit.
When a customer asks where an order is, look it up first, then reply in brand voice.
Yuna finds opportunities in real customer conversations. Nothing updates knowledge or skills until you approve it.
↳Yes, Brazil shipping usually takes 7-12 business days.
Human answered·1↳Auto-detect VIP customers · Detect VIP status and prompt agents to follow up first.
Human answered·1·L2 risk↳Opened items can be returned within 30 days if the package is complete.
Human answered·1↳If DHL has not updated for over 72 hours, check the order first, then hand off for compensation review.
You corrected·1·L2 riskManage how Agents handle customer-initiated requests. Skills affect live conversations only after simulation and publishing.
2 items| Skill | Status | Use case | Last 7 days | Actions |
|---|---|---|---|---|
| Order tracking lookupWhen a customer asks where an order is, look it up first, then reply in brand voice. | Active | Lookup | Used 12 times92% success | |
| Auto-detect VIP customersDetect VIP status and prompt agents to follow up first. | Testing | Triage | Not used yet |
Conversation volume, response efficiency, and AI service quality at a glance.
Last 7 daysLast 30 daysLast 90 days| Dimension | Conversations | Messages | Avg. first response | Resolved | Share |
|---|---|---|---|---|---|
| Web Widget | 712 | 4,128 | 38s | 496 | 60% |
| Telegram | 284 | 1,620 | 45s | 219 | 24% |
| 198 | 980 | 3m 12s | 146 | 17% |
Language & timezone
Local time 16:30 · UTC+00:00
Appearance
Account actions
Danger zone
The workspace is disabled immediately. Ops can restore it within 2 hours; after that it is permanently deleted.
Delete workspaceAgent identity
PNG, JPG, or WebP up to 5 MB
Project identity
Reply entries
Create your first quick reply; agents can invoke it by typing /.
New quick replyBrand configuration
256×256 PNG / SVG recommended; you can crop to a square after uploading
Notification permission
Notification settings
Notification scenes
Test notification
Send testSend a test notification to verify everything works.
Send key
Conversation handling
Translation settings
Create a subscription to push events to your own backend for integration or data archiving.
AI Credit
Buy AI CreditCurrent cycle resets on 09/30/2026
Plan quota 10,000
Knowledge import, parsing, and offline compilation do not consume AI Credit
Record list

Subscription active
Current cycle ends 09/30/2026.
AI Credit
Pro and Enterprise support connecting your own AI in Agents. External AI has no additional charge and uses no YundaDesk AI Credits. Third-party AI fees are paid separately; managed AI continues to use Credits.
2,400 used
Current cycle resets on 09/30/2026
1,000 used
Other usage
Seats used
3Channels connected
2Yuna requests completed this month
42New conversations this month
128Messages this month
580Documents indexed
12Deliveries this month
0These workspace-wide rules apply to proactive customer outreach. They do not affect internal reports sent to you.
Cooldown is 24 hours per customer; at most 2 touches in 7 days.
Quiet hours
Contact frequency
Team members
| Member | Role | Status & skills | Last sign-in | Last used | Actions |
|---|---|---|---|---|---|
| ACAnna ChenMeanna@yundadesk.testOnline | Admin | ActiveEnglishRefundsMax 12 concurrent | Today, 09:42 | 2 min ago | |
| LWLeo Wangleo@yundadesk.testOnline | Agent | ActiveEnglishAfter-salesMax 6 concurrent | Yesterday, 18:20 | Today, 10:06 | |
| MGMaria Garciamaria@yundadesk.testOffline | Agent | PendingSpanishPre-salesMax 8 concurrent | — | — |
Could not establish a secure connection. If the site is HTTP-only, crawl it again with an http:// URL
| Document | Size | Tags | Updated | Status | Actions |
|---|---|---|---|---|---|
| demo-owner/support-guide 2 | — | — | Updated Jun 3, 2026, 4:00 PM | Healthy |
| Document | Size | Tags | Updated | Status | Actions |
|---|---|---|---|---|---|
| https://tgo.ai/help/shipping· 47 pages | — | — | — | Searchable | ⋯ |
| Support knowledge · Notion · 128 docs | — | — | Last synced 2026/07/29 03:10 | Retrievable | ⋯ |
| Brazil shipping timelines | — | Updated Jun 3, 2026, 4:00 PM | Searchable | ⋯ | |
| EU customs & duties | — | Updated Jun 4, 2026, 2:30 PM | Indexing | ⋯ | |
| 30-day refund policy | — | Updated Jun 2, 2026, 8:00 PM | Searchable | ⋯ |
| Document | Size | Tags | Updated | Status | Actions |
|---|---|---|---|---|---|
| Brazil shipping timelineshttps://tgo.ai/help/shipping/brazil | — | shipping | 10/04 | Searchable | ⋯ |
| Load more | |||||
| Document | Size | Tags | Updated | Status | Actions |
|---|---|---|---|---|---|
| Brazil shipping policy | — | Updated Jun 3, 2026, 4:05 PM | Waiting to index | ⋯ | |
| EU customs & duties | — | Updated Jun 4, 2026, 2:30 PM | Indexing | ⋯ | |
| Brazil shipping timelines | — | Updated Jun 3, 2026, 4:00 PM | Searchable | ⋯ | |
| Import restrictions PDFError: The PDF contains no readable text | — | Updated Jun 7, 2026, 4:30 PM | Indexing failed | ⋯ |
| Document | Size | Tags | Updated | Status | Actions |
|---|---|---|---|---|---|
| https://tgo.ai/help/shipping · 0 pagesProcessed 3 / Discovered 50 pages | — | — | Crawl job running | Crawling | ⋯ |
| https://tgo.ai/help/shipping · 47 pagesLast sync succeeded · 47 pages | — | — | Last synced Jun 20, 2026, 4:00 PM | Searchable | ⋯ |
| Customer question | Standard answer | Source | Tags | Status | Actions |
|---|---|---|---|---|---|
| Do you ship to Brazil? | Shipping to Brazil usually takes 7-12 business days. | Compiled · Brazil shipping policy | Enabled | ⋯ |
Stop syncing and keep this source's 2 documents. Existing content can still be used in AI replies.
Stop syncing and move this source's 2 documents to Trash. They will no longer be used in AI replies. The remote repository and other sources will not be changed.
Separate with commas; * and ** are supported. Exclusions win. Select at least one file type.
2 files to import
- help/shipping.md
- help/returns.md
No files match. Adjust the selection or filters and preview again.
Connect Notion, choose the pages to sync, and import documents automatically.
Connect NotionNotion is not connected yet
Test the saved reception setup as a visitor without creating a real conversation.
What needs my attention today?
Three things: 2 learning suggestions to review and 1 outreach waiting for your confirmation. Yesterday's auto-resolution hit 84%, up 3 points.
Support Agent · External AI
External setup
Edit external AI- Connection
- OpenAI Chat Completions
- Endpoint
- https://api.example.com/v1
- Key
- Configured · DEMO-REDACTED
Maintain persona, knowledge, and reply logic in your external service. Use “Edit external AI” to update the endpoint or key, and the preview on the right to test replies.
External AI only handles auto-replies. It is not written into YundaDesk AI learnings and does not generate AI skills.
Your plan does not include external AI, so external reception is paused and its configuration is retained. Upgrade to Pro to resume, or create a managed Agent and reassign its channels.
Choose which tools and actions this Agent may call. These switches never replace runtime authorization, approval, or audit.
When a customer asks where an order is, look it up first, then reply in brand voice.
Both visitors and agents will see this name
Online queue, no-agents-online, and agent-joined notices apply only to messaging channels. Email channels do not send these three reception messages.
This Agent can only search explicitly selected content. Its reception focus never grants knowledge access.
Knowledge searchCould not generate suggested questions. Please regenerate them.
Selected channels are handled by the current AI. Channels not selected here will not use AI replies.
SelectThe first successful suggestions are kept without editing or saving first. Fewer than three questions, or none, may appear when suitable suggestions are limited.
Suggested questions help visitors start a conversation. A Website Widget can use its reception Agent's automatic suggestions or a custom list you save.
Manage Widget suggested questions#
Open AI Agents, select the built-in Agent receiving website conversations, then choose Suggested questions:
- Once the first suggestions are ready, the list is kept automatically. You do not need to edit and save it first, and reopening the page shows the same list. Fewer questions, or none, may appear when there are not enough suitable suggestions.
- Select a chip under Knowledge recommendations, or select Add question to write your own.
- Use the pencil to edit, the trash icon to remove, and the left handle to reorder. Select Apply question after editing a row.
- Try the questions in Debug preview on the right, then select Save at the top. Discard changes restores the last saved list.
In custom mode, Restore automatic appears to the left of Add question. Select it and save to use automatic suggestions again. Websites received by the same Agent share the list, while each website keeps its own display switches.
Custom questions define only the opening text. Clicking one asks the AI to answer normally, so check the answers in Debug preview before saving. Questions follow the Widget interface language; when translation is unavailable, text from another language is not shown.
Where suggestions come from#
Automatic suggestions use the Agent's available knowledge. A business overview, core products or services, intended audience, and getting-started guidance help produce useful opening questions.
Review the results. Check that each question relates to your core business, has a clear answer in the knowledge, and does not first require an order number or other personal information.
Where they appear#
Open Channels, select the Website Widget, and expand Display settings:
- Home: Shows the list between Chat with support and Contact us.
- Empty chats: Shows the same questions below the first conversation greeting. They disappear after the visitor sends a message.
In the Agent’s Suggested questions header, select Configure display to the left of Add question. If one website is bound, its settings open directly. If several websites are bound, choose a channel from the menu first.
The switches are independent. Select Manage beside Suggested questions to open the corresponding Agent.
Visitors see questions only from the enabled built-in Agent bound to the channel. Automatic suggestions also require knowledge retrieval. Questions are hidden during human reception, when no Agent is bound, or when the Agent is unavailable.
Use the Agent's Debug preview to try your edits. Questions under Knowledges → Test knowledge help check answers and do not change the visitor-facing list.
When they update#
After updated knowledge finishes processing, reopen or refresh the suggested questions page to see new candidates. New candidates do not automatically replace the questions already displayed.
Knowledge updates do not overwrite your saved display list. It applies on the next Widget load, while the candidates below continue to follow knowledge changes.
Regenerate suggested questions#
Select Regenerate suggested questions beside Knowledge recommendations in the Agent's Suggested questions section. You need permission to manage knowledge, and knowledge retrieval must be enabled for the Agent.
While the button shows Generating…, refreshing or leaving the page does not interrupt generation. Return to see its status and results. If the material has not changed, previous candidates remain visible during generation and after a failed attempt. You can try again after a failure.
Regeneration updates only the candidates below and keeps the current display list. Select or edit suitable questions, then select Save at the top to apply them.
If your website has changed, sync it in Knowledges first. Regeneration uses the currently saved material; it does not fetch webpages again or guarantee different questions each time.
If the suggestions are not suitable#
- Adjust the display list: Add candidates, then edit, remove, or reorder them. You can also write your own questions and save.
- Improve automatic suggestions: Update the Agent's business overview, core products, intended audience, or getting-started guidance. Wait for processing to finish, then regenerate.
- Still unsuitable: Edit and save your display list directly instead of repeatedly regenerating.
Restore automatic replaces your custom display list. Regenerate suggested questions refreshes the candidates; the two actions serve different purposes.
