AI & UX

Q: Which widget am I editing?

The widget whose card is selected at the top of Configure › AI & UX. Its id is also in the page address as ?widgetId=, and the same widget stays selected when you open User Interface or Embed.

Q: Why was my persona refused?

StrikeUp reviews every changed persona text on Configure › AI & UX. It refuses a text over its limit, one that tries to override the assistant’s own rules, one that looks like it contains a password, key or token, and one with [option: …] or [handoff] markers. The error message names the field and the reason.

Q: My persona didn’t take effect on a new widget. Why?

On a brand-new widget, the first Save Persona on Configure › AI & UX saves the widget and creates its agents, and the persona may not reach them yet. Select Save Persona a second time; the texts are then applied to every agent.

Q: Can I write different instructions for each agent?

Not on this page. Save Persona on Configure › AI & UX applies the same tone, behaviour and branding texts to every agent of the selected widget. For instructions that apply to one kind of request, write a skill on Configure › Skills.

Q: Why is there no Skills section?

The Skills section on Configure › AI & UX appears only once your workspace has at least one skill. Create a skill on Configure › Skills, or accept a suggested one on Configure › Suggestions, and the section appears.

Q: Why is there no Human Handoff section?

The Human Handoff section appears only when the helpdesk integrations feature is switched on for your workspace. When it is on, you also need a helpdesk connected on Configure › Integrations before you can pick it here.

Q: What is the difference between Easy, Normal and Hard?

They set how readily the assistant suggests a person. Easy suggests a handoff at the first mention of wanting human help, Normal helps first and suggests it if the visitor asks again or stays unsatisfied, and Hard tries every option and suggests it only if nothing works or the visitor insists.

For: Administrators and support leads who shape how a widget's assistant behavesRequires: Any signed-in user can view; saving needs the Admin role; human handoff needs the helpdesk integrations feature, and skills appear once your workspace has oneScreenshots show sample data

The AI & UX page decides how one widget’s assistant talks and what it may do beyond calling your API: its tone and instructions, the skills it can use, when it offers a person instead, and whether it can send emails. Each setting applies to the widget you select at the top, so two widgets can behave differently. Open it from ConfigureAI & UX. The whole page, including its title, section names and buttons, is in English in every language; this guide quotes those labels as the page shows them.

What you can do here

Pick the widget to configure. A row of cards lists your widgets; the selected one is the widget every section below edits. Choose or create a widget.
Give the assistant a persona. A tone of voice and behaviour and branding instructions are added to every agent of the widget. They cannot relax its safety, approval or guardrail rules. Set the persona.
Choose its skills. Switch each of your workspace’s skills on or off for this widget; the change applies at once. Choose the widget’s skills · Skills.
Let visitors reach a person. Show an Invite an agent button and decide how readily the assistant suggests a handoff to your helpdesk. Set up human handoff · Human handoff.
Let the assistant send emails. Allow it to email reports and data exports through your email channels, and pick the default sender. Turn on AI email notifications · Notifications.

The screen at a glance

The AI & UX page under Configure, subtitled "Configure agents, skills, handoff, and email notifications per widget." Two widget cards are shown; "Demo Ltd Support chat" (custom-trigger, toolkits Demo Billing API and Demo Accounts API) is selected, next to "Demo Ltd Help Centre widget" and a dashed "New widget" card. Below, the Persona section ("Tone, behavior and branding instructions applied to all agents") is filled: Tone of voice "Friendly, clear and concise. Plain English, no jargon.", behaviour and branding instructions. The blue "Save Persona" button is highlighted; it saves the three texts for every agent of this widget. — Pick a widget card, then set the tone, behaviour and branding its assistant uses and select Save Persona.

Region	What it shows	What you do there
Widget cards, top	One card per widget with its display type and toolkits, plus a dashed `New widget` card	Pick the widget to edit, create a new one, delete one
`Persona`	Tone of voice, behaviour instructions and branding instructions	Write them and select `Save Persona`
`Skills`	One card per skill in your workspace, with a switch	Switch skills on or off for this widget
`Human Handoff`	The handoff button switch, handoff sensitivity, integration and inbox or team	Configure handoff and select `Save Handoff Settings`
`Email Notifications`	The AI email switch and the connected mail channels	Link channels, choose the default, select `Save Notification Settings`

Each section is collapsible. Until you pick a widget, the page reads Select a widget above to configure its AI settings.

The selected widget card is the one you are editing. The choice is kept in the page address (?widgetId=…) and carries over when you switch to User Interface or Embed.

To create a widget:

Select the dashed New widget card. The Create New Widget Configuration dialog opens.
Enter Widget Name *, and optionally a description.
Choose a display type, pop-up or sidebar, and whether visitors must sign in. Sign-in is on by default.
Select the toolkits the widget may use. You can select up to 5 toolkits, although the dialog’s labels speak of one. Only active toolkits are listed, and tool names must be unique across the toolkits you pick.
Optionally select knowledge base documents and web sources, then select Create Widget.

The new widget gets its own embed token, a set of default agents and two starter shortcut cards on its welcome screen, Get started and Build me a report. It is selected automatically a moment later. Set its colours, size and other display settings on User Interface afterwards; the dialog does not keep them. More on choosing toolkits and knowledge is on Embed.

To delete a widget, open the ⋮ menu on its card, select Delete and confirm Delete this widget?. Every card shows a green Active badge whatever its settings.

Set the persona

The persona is the voice of the widget’s assistant. Your three texts are added to the instructions of every agent the widget has, under a heading that tells the assistant they cannot relax its safety, approval or guardrail rules.

Pick the widget card at the top.
In Persona, fill in Tone of voice (optional), for example Friendly, clear and concise. Plain English, no jargon.
Add Behavior instructions (optional), for example Always confirm the invoice number before discussing an invoice.
Add Branding instructions (optional), for example Refer to the product as Demo Ltd, never as “the platform”.
Select Save Persona. The message Widget saved successfully confirms the save.

StrikeUp reviews each changed text before saving it. A text is refused, with Request failed and the reason, when it:

Reason shown	What triggered it
“This is 4210 characters; the limit is 4000.”	A text longer than 4,000 characters. The tone of voice has its own limit of 500 characters
“This text tries to override or disable the assistant’s own rules…”	Wording such as “ignore your previous instructions”
“This text looks like it contains a password, key or token…”	Something that looks like a credential. It would be sent to the AI model in every turn
“Remove the [option: …] or [handoff] markers…”	Control markers the chat widget acts on

The reason starts with the field it concerns, such as Behavior instructions:. A text you did not change is not reviewed again, and when the review itself cannot run, the text is refused rather than saved.

Do	Don’t
Describe behaviour the visitor will notice: tone, what to confirm, what to avoid	Try to switch off approvals or guardrails; persona text cannot do that
Name your product and terms the way your customers know them	Put API keys, passwords or internal URLs in the text
Keep each instruction to one sentence	Paste long policies; use Knowledge Base for reference material

On a brand-new widget whose agents are not set up yet, the first Save Persona saves the widget and creates its agents, and the persona texts may not reach them. Select Save Persona a second time to apply them.

Choose the widget’s skills

A skill is a set of instructions the assistant follows for one kind of request, such as invoice disputes. Skills are written once for your workspace on ConfigureSkills; here you decide which ones this widget’s assistant may use.

The Skills section of the AI & UX page for "Demo Ltd Support chat", described as "Enable skills available to all agents in this widget", with a "Manage Skills →" link at the right. Three skill cards each show a name, the situation that triggers it and a switch: "Invoice disputes" and "Plan changes" are switched on; the highlighted "Usage questions" ("Activate when a customer asks how many API calls they have used or why a usage charge is high.") is switched off for this widget. Flipping a switch adds or removes that skill for this widget at once. — Each company skill has a switch per widget; changes apply immediately without a Save button.

Each card shows the skill’s name, the situation that triggers it and a switch. Flipping a switch saves at once; there is no Save button. Manage Skills → opens the skills page. The Skills section appears only when your workspace has at least one skill.

Set up human handoff

Human handoff lets a visitor continue the chat with a person in your helpdesk. The Human Handoff section appears only when the helpdesk integrations feature is switched on for your workspace, and it needs a helpdesk connected on Integrations.

The Human Handoff section of the AI & UX page ("Escalate conversations to a live agent when the AI can't resolve the issue"). The switch 'Enable "Invite an agent" button' is on. Under Handoff sensitivity three options are shown: "Easy" (first mention of wanting human support triggers a handoff suggestion), the selected and highlighted "Normal" (AI tries to help first; suggests handoff if the user asks again or remains unsatisfied) and "Hard" (AI exhausts all options first). Integration is "Help Scout — Billing support (HelpScout)", Inbox / Team is "Billing", followed by "Save Handoff Settings". — Handoff sensitivity decides how soon the assistant offers a live agent; Normal is the default.

Switch on Enable "Invite an agent" button. The widget’s chat then shows a handoff button.
Under Handoff sensitivity, pick how readily the assistant suggests a person.
Under Integration, pick the connected helpdesk. For Help Scout, pick a mailbox under Mailbox; Save Handoff Settings stays disabled until you do. For Intercom, optionally pick a team under Inbox / Team (optional).
Select Save Handoff Settings.

Sensitivity	What the assistant does
`Easy`	The first mention of wanting human support triggers a handoff suggestion
`Normal` (the default)	The assistant tries to help first, and suggests a handoff if the visitor asks again or remains unsatisfied
`Hard`	The assistant exhausts all options, and suggests a handoff only if nothing works or the visitor insists

With no helpdesk connected, the section reads No support integrations connected. Go to Integrations to connect Help Scout. The full behaviour, from the visitor’s request to the helpdesk conversation, is in Human handoff.

Turn on AI email notifications

With AI email notifications on, the assistant can email a visitor a report or a data export when they ask for one, through one of your email channels.

The Email Notifications section of the AI & UX page ("Let the AI send reports and data exports to users via email"). The switch "Enable AI email notifications" is on, with the hint "When enabled, the AI can send emails when users ask for reports or data exports". Under Connected channels, "Billing team email" (notifications@example.com · Active) is linked and marked "Default"; the Telegram channel "Demo Ltd assistant bot" (Active) is linked too, and its highlighted "Set default" button would make it the default; "Demo Ltd Slack" is listed but not linked. "Save Notification Settings" stores the choices. — Turn on AI email notifications, link the channels the assistant may send through and choose the default one.

Switch on Enable AI email notifications. It is off by default.
Under Connected channels, switch on each channel the assistant may use. The first channel you link becomes the default.
To change the sender, select Set default on another linked channel. The default one reads Default.
Select Save Notification Settings.

The channels themselves are created on ConfigureNotifications. Without any, the section reads No notification channels configured yet. Go to Notifications in the sidebar to add an SMTP channel first. A channel that is disabled there shows · Disabled and is not used.

What the assistant can send: an email with a subject, a body and, where the visitor asked for data, a CSV attachment, to at most 5 addresses at once. Who may receive it is set on each SMTP channel on ConfigureNotifications under Who the AI can email: Signed-in user only (the default) sends only to the signed-in visitor’s own confirmed address, Allowed list also allows the addresses and @domains you list, and Any address allows any address the visitor gives. An unconfirmed address is confirmed first with a 6-digit code the visitor types in the chat, and the assistant sends at most 50 emails to one customer and 1,000 for your workspace in 24 hours.

Every control, explained

Where

The row at the top of the page. Selecting a card puts its id in the page address and reloads every section.

What

One card per widget with its name, display type and toolkits. The ⋮ menu offers Delete. A preview icon, where your installation offers it, opens the widget in a new browser tab with a temporary preview visitor. The dashed New widget card opens Create New Widget Configuration.

Why

Every section on the page edits one widget, so you always see which widget you are changing.

Persona

Where

The Persona section, saved with Save Persona.

What

Tone of voice (optional), up to 500 characters; Behavior instructions (optional) and Branding instructions (optional), up to 4,000 characters each. They are added to every agent of the widget and reviewed before saving. One AI review call is made per changed text.

Why

Visitors notice tone first. A persona lets each widget sound like the product it sits in without touching the assistant’s safety rules.

Skills switches

Where

The Skills section; each switch saves the moment you flip it. Manage Skills → opens ConfigureSkills.

What

One switch per skill in your workspace. On: the assistant of this widget can use the skill when its trigger situation comes up. Off: it cannot. If a switch cannot be saved, a Request failed message appears.

Why

Different widgets serve different audiences; a billing widget needs the invoice skills, a help-centre widget may not.

Human Handoff

Where

The Human Handoff section, shown with the helpdesk integrations feature. Saved with Save Handoff Settings.

What

Enable "Invite an agent" button shows the handoff button in the chat. Handoff sensitivity is Easy, Normal (default) or Hard. Integration picks the connected helpdesk; Mailbox (Help Scout, required) or Inbox / Team (optional) (Intercom) picks the destination inside it. The chosen sensitivity is added to the assistant’s instructions while handoff is on.

Why

Some visitors need a person. The sensitivity balances solving questions in the chat against making people repeat themselves.

Email Notifications

Where

The Email Notifications section. Saved with Save Notification Settings.

What

Enable AI email notifications (off by default) gives the assistant an email tool. Under Connected channels, every channel from ConfigureNotifications is listed with a switch and Set default; linking an email channel gives the assistant an email tool, a Telegram channel a tool that sends a Telegram message to a chat ID the visitor gives, and Slack or Teams channels add nothing; the default one is marked Default. The tool is offered only when the switch is on and at least one channel is linked and enabled.

Why

Reports and exports are often too large for a chat message. Email delivers them where the visitor can keep them.

Who sees what

Part of the screen	Shown when
The page, the widget cards, `Persona` and `Email Notifications`	Always, to every signed-in user
`Skills`	Your workspace has at least one skill
`Human Handoff`	The helpdesk integrations feature is switched on for your workspace

Everyone sees the same controls, but only users with the Admin role can change them. For anyone else, creating a widget, the Save buttons and the skill switches fail with Request failed and “This action requires the Admin role.”

What you need to know

Topic	Fact
Per widget	Every setting on this page belongs to the selected widget only. Skills themselves are shared across the workspace.
Saving	Persona, handoff and email each have their own Save button. Skill switches save instantly.
Persona limits	Tone 500 characters, each instruction 4,000. Changed texts are reviewed before saving.
Safety rules win	Persona text cannot relax safety, approval or guardrail rules.
Cost	Each changed persona text costs one AI review call. Emails the assistant sends add no AI review cost.
Language	The page itself is in English in every language.

Frequently asked questions

Widgets

The widget whose card is selected at the top of ConfigureAI & UX. Its id is also in the page address as ?widgetId=, and the same widget stays selected when you open User Interface or Embed.

Between 1 and 5. The create dialog’s labels mention a single toolkit, but you can select up to 5 active toolkits there and on ConfigureEmbed. Tool names must be unique across the selected toolkits, or the widget is refused.

The Create New Widget Configuration dialog saves only the name, description, display type, sign-in choice, toolkits and knowledge. Colours, size and the other display settings start with their defaults; change them on ConfigureUser Interface.

Persona

Why was my persona refused?

StrikeUp reviews every changed persona text on ConfigureAI & UX. It refuses a text over its limit, one that tries to override the assistant’s own rules, one that looks like it contains a password, key or token, and one with [option: …] or [handoff] markers. The error message names the field and the reason.

On a brand-new widget, the first Save Persona on ConfigureAI & UX saves the widget and creates its agents, and the persona may not reach them yet. Select Save Persona a second time; the texts are then applied to every agent.

Can the persona make the assistant skip approvals or guardrails?

No. Persona texts are added under a heading telling the assistant they cannot relax its safety, approval or guardrail rules, and wording that tries to override those rules is refused when you save. Approvals are set per tool on ConfigureToolkits.

Can I write different instructions for each agent?

Not on this page. Save Persona on ConfigureAI & UX applies the same tone, behaviour and branding texts to every agent of the selected widget. For instructions that apply to one kind of request, write a skill on ConfigureSkills.

Skills, handoff and email

Why is there no Skills section?

The Skills section on ConfigureAI & UX appears only once your workspace has at least one skill. Create a skill on ConfigureSkills, or accept a suggested one on ConfigureSuggestions, and the section appears.

Why is there no Human Handoff section?

The Human Handoff section appears only when the helpdesk integrations feature is switched on for your workspace. When it is on, you also need a helpdesk connected on ConfigureIntegrations before you can pick it here.

What is the difference between Easy, Normal and Hard?

They set how readily the assistant suggests a person. Easy suggests a handoff at the first mention of wanting human help, Normal helps first and suggests it if the visitor asks again or stays unsatisfied, and Hard tries every option and suggests it only if nothing works or the visitor insists.

What do I need before the assistant can send emails?

An SMTP channel created on ConfigureNotifications. Then, on ConfigureAI & UX, switch on Enable AI email notifications for the widget, link the channel under Connected channels and select Save Notification Settings.

Whom can the assistant email?

It depends on Who the AI can email on the SMTP channel on ConfigureNotifications. With Signed-in user only, the default, only the signed-in visitor’s own address is used, and an address they have not confirmed yet first gets a 6-digit code to type into the chat. Allowed list adds the addresses and @domains you list, and Any address allows any address, up to 5 in one email. The email goes out through the widget’s default channel set on ConfigureAI & UX, unless the visitor’s customer company has its own active mail channel, which is then used instead.

Troubleshooting

Symptom	What to check
Request failed when saving the persona	Read the reason in the message. Shorten the text, remove anything that looks like a credential, or remove wording that tries to override the assistant’s rules.
The persona is saved but the assistant does not use it	On a new widget, select `Save Persona` again. Check that you edited the widget you are testing.
The page reads `Select a widget above to configure its AI settings.`	Pick a widget card. If there is none, create a widget with the `New widget` card.
No Skills section	Your workspace has no skills yet. Create one on ConfigureSkills.
No Human Handoff section	The helpdesk integrations feature is off for your workspace.
The assistant never offers to email	Check that `Enable AI email notifications` is on, a channel is linked and enabled, and `Save Notification Settings` was selected.
In the create dialog, the remaining toolkit switches are greyed out	You have selected 5 toolkits, the limit. Deselect one to pick another; the dialog’s count still reads `of 1`.
A change fails with Request failed and “This action requires the Admin role.”	Changes on this page need the Admin role. Ask an Admin to make the change, or to give you the Admin role under Settings › Users.

AI & UX

What you can do here

The screen at a glance

Choose or create a widget

Set the persona

Choose the widget’s skills

Set up human handoff

Turn on AI email notifications

Every control, explained

Widget cards

Persona

Skills switches

Human Handoff

Email Notifications

Who sees what

What you need to know

Frequently asked questions

Widgets

Which widget am I editing?

How many toolkits can a new widget have?

Why did my new widget ignore the colours I expected?

Persona

Why was my persona refused?

My persona didn’t take effect on a new widget. Why?

Can the persona make the assistant skip approvals or guardrails?

Can I write different instructions for each agent?

Skills, handoff and email

Why is there no Skills section?

Why is there no Human Handoff section?

What is the difference between Easy, Normal and Hard?

What do I need before the assistant can send emails?

Whom can the assistant email?

Troubleshooting

AI & UX

What you can do here

The screen at a glance

Choose or create a widget

Set the persona

Choose the widget’s skills

Set up human handoff

Turn on AI email notifications

Every control, explained

Widget cards

Persona

Skills switches

Human Handoff

Email Notifications

Who sees what

What you need to know

Frequently asked questions

Widgets

Which widget am I editing?

How many toolkits can a new widget have?

Why did my new widget ignore the colours I expected?

Persona

Why was my persona refused?

My persona didn’t take effect on a new widget. Why?

Can the persona make the assistant skip approvals or guardrails?

Can I write different instructions for each agent?

Skills, handoff and email

Why is there no Skills section?

Why is there no Human Handoff section?

What is the difference between Easy, Normal and Hard?

What do I need before the assistant can send emails?

Whom can the assistant email?

Troubleshooting

Related