AI support chatbot

VROF chatbots answer only from content published on the site.

It learns from your site's content and answers only within that scope. For questions outside it, it does not make up an answer and forwards the question to an inquiry.

• Automatic masking of personal information

  Six types of data (resident registration numbers, account numbers, card numbers, passwords, phone numbers, and email addresses) are masked twice: before logs are stored and before anything is sent to the AI model. Only masked values are sent externally.

• Conversations deleted automatically after 90 days

  Conversation logs are kept to understand the types of questions asked, and a daily cleanup job deletes logs older than 90 days.

• Handover to staff

  When a visitor asks to speak with a person, a notification is sent to the staff email along with the page visited and the conversation.

• Support for 5 languages

  Korean, English, Japanese, Chinese, and French are supported in a single setup, and the guide's name differs by language (브이 · Vee · ブイ · 小V · Vee).

[Chat] Handover request · vrof

Site   : vrof
Page   : /service/chatbot
Masked : 1

── How to reply ──
This email contains no visitor contact details. Phone numbers and email addresses are removed from the conversation before storage, and the visitor was informed that contact details are not stored.

── Conversation so far ──
Visitor: I would like to discuss a quote with a person
Vee: Please leave an inquiry and someone will follow up

The same engine also runs an AI saju consultation service.

Beyond FAQ responses, VROF's chatbot engine handles chart-based readings, crisis response, and privacy protection in the live “AI Fortune Teller” service.

• Answers based on structured data

  Rather than stopping at a fixed FAQ, it generates answers from structured data such as a computed saju chart.

• Referral to professional counseling on crisis signals

  When a user's expression of risk is detected, it stops the AI response and connects the user to a professional counseling line.

• On-device processing of sensitive information

  Sensitive source information such as the date and time of birth stays on the user's device, and only what is needed is sent to the server.

Visit AI Fortune Teller →

The guide's expression shows the current status.

Visitors can tell what is happening from the guide's expression, even before reading the answer.

• Expression while waiting

  When a question is sent, the expression changes to thinking and stays until the first character of the answer appears.

• Expression when it cannot answer

  When there is no answer on the site and the question is forwarded to an inquiry, the expression changes first, so the message comes across more gently.

• Five expressions from a single image

  A single image with five expressions placed side by side is used, so expressions switch instantly without additional downloads.

• Five expressions for each conversation situation

  Greeting (when the window opens) · Thinking (while waiting for an answer) · Answering (when it answers) · Unable to answer (when forwarding to an inquiry) · Handover (when Leave an inquiry is clicked). Past messages keep the expression from that moment.

• Greeting
• Thinking
• Answering
• Unable to answer
• Handover

It answers real visitor questions in about 6 seconds on average.

It answers real visitor questions every day on this site and eet.kr. The figures below count only real visitor conversations, excluding test calls.

• Per conversation KRW 8.83 Based on 12 human conversations
• Average response 6.49s Based on the same 12
• Failures 0 Out of 20 calls

One server runs multiple sites, with training data and conversation logs managed separately for each site.

An admin screen for analyzing visitor questions is included.

The admin screen comes with the chatbot. Below is the actual screen in use; the figures shown are sample data used to illustrate the distribution.

• Unanswered questions shown first

  Unanswered questions are shown first, so you can immediately see what to add to the training data and which inquiries staff should handle. The three entries on the screen above are real records: two were unanswered, and one was an out-of-scope question that was declined.

• Sample size shown with every figure

  Every percentage is shown with its base count. For example, “2.8% failures” means 1 out of 36 calls.

• Filter by period and site

  You can view Today · 7 days · 30 days · All, or a custom period. The selected period is saved in the URL, so the same screen appears after a refresh or when the link is shared.

• Read-only

  Neither the screen nor the server has any function to delete or modify records, which keeps the records reliable.

This screen is where VROF reviews conversations on its own site. Conversation analysis for client sites is not yet offered.

It can be introduced without any development work on your side.

The process after your inquiry is as follows and can be used as is for internal approval.

• ① Answer scope list

  We compile likely questions based on your site's content and divide them into those the chatbot will answer and those staff will answer. You can confirm the scope before development.

• ② Training data build and testing

  We build training data from your site's content and check the answers with real questions. Wrong answers are fixed by refining the training data and testing again.

• ③ Site integration

  VROF installs the widget to fit your site's structure, with no work required from your development team. The two sites currently in operation were integrated on Astro and React, respectively. VROF manages the server, model, and usage costs.

• Training data updates when the site changes

  Let us know when your site content changes, and we will rebuild the training data. Wrong answers are corrected by finding the question on the admin screen, which is why unanswered questions are shown first.

• How it differs from scenario-based chatbots

  Chatbots built on predefined buttons and branches struggle with unexpected questions. VROF's chatbot finds answers in your site's content, so it can answer questions no one planned for, and it does not answer what is not on the site.

Support chatbot pricing.

The three plans below and the monthly operating fee (VAT excluded) can be used as is in internal approval documents.

• Standard

  KRW 490,000

  One language · Installed with a single line of script · Up to 30 pages of site content as training data · Automatic masking of six types of personal information · Conversations deleted automatically after 90 days · Handover to staff by email when it cannot answer

• Deluxe

  KRW 990,000

  Everything in Standard, plus — Five-language set (Korean, English, Japanese, Chinese, French) · Five facial expressions · Read-only admin screen (metrics by period · unanswered questions · defended questions)

• Premium

  KRW 1,990,000

  Everything in Deluxe, plus — Widget integration fitted to your site's framework · Adjustments to match your site design

The monthly operating fee is KRW 30,000–50,000 and includes 1,000 conversations, with KRW 30 per conversation beyond that. It covers the server, model, personal information masking, 90-day deletion, and the admin screen. New features and screen changes are not included.

Rebuilding training data after major site content changes, or re-integrating the widget after a site redesign, is billed separately. The scope of coverage is specified in the initial contract.

Related services

The chatbot answers based on content published on the site and does not answer questions about anything not published. Work that takes over staff judgment is handled by our AI adoption service, and moving repetitive internal work into software by our Workflow automation service.

Talk to VROF about a support chatbot.

Send us your site URL. Before development, we will list likely questions and divide them into those the chatbot will answer and those staff will answer.

Reply within 1 business day