Feature Settings

Conversation platform feature settings: basic features, AI interaction features, Hook processing markers, tool display mode, question option display mode, AI awareness settings (system time, sender na

Feature Settings has the most switches on the conversation platform settings page. It determines what users can do and see in a conversation, and what the AI knows when it responds. Web and API platforms have all six sections. The internal platform does not have “Question Option Display Mode”; messaging platforms such as LINE and Microsoft Teams have only Enable OCR and “AI Awareness Settings.”

These settings belong to each conversation platform, not to the AI agent. If the same agent is connected to both Web and LINE, configure each platform separately. For the overall layout and how to access the settings page, see Conversation Platform Settings Overview.

1. When to Use This

A clinic uses an AI agent on LINE to answer appointment questions. A patient asks, “Can I still book an appointment this afternoon?” The agent responds by asking, “What is today's date?” An administrator opens Feature Settings for the LINE platform, enables Use System Time in Responses, leaves the time zone set to Asia/Taipei, and clicks Save. When asked again, the agent answers directly: “Today is Friday. The afternoon clinic starts at 14:00, and appointments are still available.” The agent's role instructions have not changed at all.

2. Basic Features

Web platform Feature Settings: the last row of Basic Features contains Enable Conversation Handoff and Show Context Usage

Switch	Description
Enabled	When turned off, the platform stops serving users, who can no longer send messages from its conversation page
Allow File Uploads	Users can upload images or files for AI analysis during a conversation
Allow Conversation Sharing	Users can generate a link to share a conversation
Allow Font Size Changes	The conversation window provides a font size control
Enable Conversation Handoff	When MaiAgent has enabled conversation handoff globally, Web Chat on this platform shows a handoff prompt bar once context usage reaches 80% and can generate a handoff summary card. The global switch is not in the admin console; contact MaiAgent. Turning this switch on has no visible effect unless the global feature is enabled. Off by default
Show Context Usage	Web Chat on this platform shows the context usage percentage below the input box. This is independent of conversation handoff; either feature can be enabled on its own. Off by default

Enable Conversation Handoff and Show Context Usage appear only on Web and API platforms. The internal platform does not show these switches. MaiGPT needs no configuration: context usage is always shown, and conversation handoff depends only on the global switch. For their effects during a conversation, see Conversation Handoff.

3. AI Interaction Features

Switch	Description
Show Cited Documents	List the knowledge base sources referenced in an AI response; users can click a citation to view the document excerpt
Allow Downloading Cited Documents	Users can download cited files
Enable OCR	Recognize text in uploaded images so the AI can read screenshots or scans
Show Thinking Process	Show the AI's reasoning process in the conversation
Allow Users to Adjust Reasoning Effort	Let users adjust the AI's reasoning depth in the conversation window

Conversation after Show Cited Documents and Allow Downloading Cited Documents are enabled

Choose whether to expose citation information based on the use case: public customer support can show all citations to build trust; internal support can disable downloads to protect confidential documents; educational institutions can show learning resources for students to consult.

4. Hook Processing Markers

When Show Hook Processing Markers is enabled, the conversation displays a marker when a message is processed by a Hook, such as one that masks sensitive information. You can customize the marker text. For Hook configuration, see Hook Message Interception.

5. Tool Display Mode

Choose one of three cards to control how much of an AI tool call (MCP or API) appears in the conversation window:

Mode	Effect in the conversation
Full	Show the tool name; users can expand it to see input parameters and the response
Compact	Show only the tool name and execution status
Hidden	Show no tool execution information

Choose one of three Tool Display Modes

The “Display Text” tab for an individual tool also has a display mode. It can follow the conversation platform setting or specify Full, Compact, or Hidden for that tool. See Customize Tool Message Display. The text and waiting animation shown while a tool runs are configured in Response Waiting Style.

6. Question Option Display Mode

After an AI agent is configured with the built-in Agent UI Question Tool, it can present a question to users during a conversation. For example, it can ask “Which topic would you like to learn about?” and offer several options, then continue based on the selected answer. This section controls how those options appear in Web Chat. Choose one of two cards, then click Save for the whole page:

Mode	Effect in the conversation window
Floating Buttons	Options appear as a row of buttons above the input box, which remains usable. If a user types and sends a message, the options are ignored
Option Panel	A panel of options covers the input area. The input box returns after the user answers or closes the panel

Question Option Display Mode: both cards show thumbnail previews; the selection is saved with the rest of the page

A customer support manager's agent asks visitors to choose among “Product Features,” “Pricing Plans,” and “Technical Support” at the start of a conversation. After launch, a visitor says, “I want to type my question, but the option panel blocks the input box.” The manager opens Feature Settings for that Web platform, selects Floating Buttons under Question Option Display Mode, and clicks Save. When visitors reopen the conversation window, the options appear in a row above the input box. They can click an option or type without first closing a panel.

“Option Panel” is the default for both new and existing platforms. To use Floating Buttons, switch modes manually. This setting affects Web Chat on Web and API platforms only. Test conversations on the internal platform always use an option panel, so this section does not appear on its settings page.

Have the Agent Ask Questions: Role Instruction Examples

To have an agent ask questions during a conversation, go to AI Features → AI Agents, then configure the built-in Agent UI Question Tool on the Tools tab. The tool tells the AI which fields each question supports. Role instructions only need to specify three things: when to ask, what to ask, and whether to allow free-form input. There is no “Allow Other” switch in the admin console; role instructions determine this, and the AI includes the setting each time it asks a question.

Fields available for each question:

Field	Purpose	If omitted
label	Question text	Required
options	List of options; 2 to 5 recommended	Required
multi_select	Allow multiple selections	Single selection
allow_other	Allow free-form “Other” input: the option panel adds an “Other” input box, while Floating Buttons changes the input placeholder to “Enter another answer”	Users can only choose an option
skippable	Set to false to require an answer: the option panel hides “Skip,” and the Esc key does nothing	Can be skipped

Example 1: Clarify an ambiguous request (single selection)

Role instruction excerpt

If a user's question could refer to several plans and you cannot tell which one to answer about, use the Agent UI Question Tool to clarify instead of guessing.
For example, if the user says “I'd like to learn about the plans,” ask “Which plan would you like to learn about?” with these options: Starter, Professional, Enterprise.
Ask at most 2 questions at a time, with 2 to 5 options each. Once you receive an answer, respond directly without asking the same question again.

Example 2: Allow “Other” and multiple selections

Role instruction excerpt

When asking about the user's industry, offer Finance, Retail, and Manufacturing, and allow them to enter “Other” (set allow_other to true).
When asking which features they want to learn about, allow multiple selections (set multi_select to true) and offer Knowledge Base, Workflows, Multilingual Support, and API Integration.

Example 3: Ask only once at the start of a conversation

Role instruction excerpt

In the first response of a conversation, use the Agent UI Question Tool to ask “Which topic would you like to learn about?” with these options: Product Features, Pricing Plans, Technical Support.
Do not ask another question proactively unless the user says “Choose again.”

Example 4: Required question (cannot be skipped)

Role instruction excerpt

Every quiz question must be answered (set skippable to false).
If the user types instead of selecting an option, or closes the question, explain in one sentence that an answer is required, then ask the same question again.

skippable: false affects only the “Skip” button and Esc key in the option panel. Users can still close the panel with ✕. Floating Buttons has no Skip button, and typing directly ignores the options. To make a question truly required, state in the role instructions, as in Example 4, that the agent must ask it again if the user does not answer.

Questions generated by the AI from these instructions look like this (assembled automatically by the AI for reference; do not put this in the role instructions):

First question from Example 2

{ "id": "industry", "label": "Your industry?", "options": ["Finance", "Retail", "Manufacturing"], "allow_other": true }

After a user answers, Web Chat sends the response in the format “Q: question” and “A: answer.” Multiple selections are joined with “、”; a free-form “Other” answer uses the same format. To route responses based on the answer, use this format in the role instructions.

Do not write “Click Skip” or “Close the panel” in role instructions or question text: Floating Buttons has neither button. Sending a typed message ignores the options; if allow_other is enabled for the question, the typed message is treated as an “Other” answer.

7. AI Awareness Settings

AI Awareness Settings let an AI agent know the current time, whom it is speaking with, and where that person is. When enabled, the system automatically adds the corresponding information to the AI's context before every response. You do not need to change role instructions or connect another tool. Each platform has its own three switches. The Web, internal, and API platforms created automatically with an agent have all three off by default.

Switch	What the AI knows	Suitable scenarios	Available on
Use System Time in Responses	Current date, time, day of the week, and time zone	“What's today's date?”, “What time is it?”, “Are you still open?”, “What day of the week will it be in three days?”	All platforms
Use Sender Name in Responses	The other person's name	Address customers by name; personalize greetings	All platforms
Use Location in Responses	Coordinates shared by the user	“Which store is closest to me?”, “Is there a service location nearby?”	Web, internal, API, LINE

Use System Time in Responses

Scroll to AI Awareness Settings and enable Use System Time in Responses. On Web and internal platforms, “Automatic Time Zone Detection” appears. The time zone comes from the user's browser, so no manual setting is needed.

Web platform: when enabled, Automatic Time Zone Detection appears and uses the user's browser time zone

On API and messaging platforms such as LINE, enabling this setting adds a Time Zone dropdown. The Current Time in that time zone appears below it for confirmation. The default is Asia/Taipei (Taipei time, UTC+8); search to switch to another time zone.

API platform: select a time zone manually; the current time in that zone appears below

How the Time Zone Is Determined

The injected time is converted to the user's time zone. Its source depends on the platform:

Platform	Time zone source	Manual setting needed?
Web, internal	Automatically taken from the user's browser and included with every message	No; the interface shows “Automatic Time Zone Detection”
API	The caller includes timezone in the message metadata (IANA format, such as America/New_York); otherwise, the platform's Time Zone setting is used	Recommended as a default when no time zone is supplied
Messaging platforms such as LINE, Microsoft Teams, Slack, and Email	Always use the platform's Time Zone setting	Yes; for a platform serving mostly overseas users, choose their time zone

Messaging platforms cannot obtain each user's time zone. If one LINE platform serves users in several countries, the AI can answer using only one fixed time zone. Consider creating a separate platform for each region.

Example of including a time zone in a message sent through the API:

{
  "message": {
    "content": "What time is it?",
    "metadata": { "timezone": "America/New_York" }
  }
}

If the time zone string is invalid, the API rejects the request and returns a validation error.

Use Sender Name in Responses

When enabled, the AI knows the other person's name and uses it when appropriate. The name comes from the contact on this conversation: the name shown under Customer Conversations → Contacts. On messaging platforms, the contact is created automatically from the user's display name; on Web platforms, the name comes from the logged-in user or the information entered in the pre-chat form. If the contact has no name, the AI responds as usual.

Use Location in Responses

When enabled, the AI uses the user's coordinates to give location-aware answers, such as recommending the nearest store. For Web Chat, the conversation window requests location permission from the browser and includes the coordinates with the message. On API platforms, the caller includes latitude and longitude in the message metadata. This switch is currently available on Web, internal, API, and LINE platforms.

8. Frequently Asked Questions

Q: I enabled “Use System Time in Responses,” but the AI still gives the wrong date. Why?

First, confirm that you enabled the switch on the platform the user actually uses (for example, LINE rather than Web if the problem occurs on LINE). Then check the Time Zone setting on the messaging platform.

Q: Do I need to add “Today is [date]” to the role instructions?

No, and it is not recommended. The system supplies the time automatically for every response. Hard-coding a date in the role instructions could give the AI two conflicting dates.

Q: Does the AI call an external service to check the time?

No. The time comes from the platform server's clock. It is converted to the applicable time zone and given directly to the AI, without calling an external API.

Q: Is daylight saving time handled automatically?

Yes. The time zone is set with an IANA name (such as America/New_York), and the system converts the time automatically when daylight saving time changes.

Q: Is this time zone the same as the service hours time zone in Human Handoff Settings?

They are separate settings. This time zone is for AI responses. The service hours in Human Handoff Settings have their own time zone, which determines when human support agents are on duty.

Q: If I hide tool cards with Tool Display Mode, will the AI still use tools?

Yes. “Hidden” only hides the tool execution area in the conversation; it does not disable tools. To stop the AI from calling a tool, remove it from the AI agent's Tools configuration.