Chapter 13: Concierge Builder

Chapter 13: Concierge Builder — ThinkNavi User Manual

Chapter 13: Concierge Builder

A feature for automatically generating AI chat widgets from your website or documents and embedding them on external sites. Access by clicking “Concierge” in the sidebar.

13.1 What Is Concierge?

Concierge is an AI chatbot that learns from your documents. When visitors ask questions, it responds based on document content.

Use Cases:

Automated responses for company help pages and FAQs
Interactive product manual guides
Internal knowledge base search support
Q&A for onboarding materials

13.2 Concierge List Screen

The first screen displayed.

Display:

“Create New” button
Card list of created concierges

Card Information:

Element	Description
Concierge Name	Configured name
Status Badge	draft / processing / ready / active
Source Type	URL / File
Created Date	Creation date and time
Delete Button	Trash icon

Operations:

Click “Create New” to start the creation wizard
Click an existing card to enter the edit screen
Trash icon to delete (with confirmation dialog)

13.3 Step 1: Source Selection

Specify the data source for the concierge to learn from.

Concierge Name Input

Enter a name in the text input field.

Examples: “Help Desk AI”, “Product Guide”, “Internal Knowledge”

Source Type Selection

Choose from two buttons:

“Website” Button — Collect Content from URL

Enter a starting URL (e.g., https://example.com/docs/)
Set maximum page count (1-500, default 100)
Click “Start Collection”

How Crawling Works:

Fetches the starting URL page
Collects URLs within the same domain/path from links on the page
Extracts body text using Readability
Crawls recursively until reaching the maximum page count

Crawl Results Display:

List of collected pages (title, path, text volume)
Remove unwanted pages with the trash icon

Notes:

Pages blocked by robots.txt cannot be collected
Content dynamically generated by JavaScript may not be retrievable
Pages requiring login cannot be collected

“File” Button — File Upload

Select files with the “Select Files” button (multiple selection supported)
After upload, file list is displayed

Supported File Formats:

Format	Extension	Max Size	Processing
PDF	.pdf	20MB	Text extraction (image-based PDFs not supported)
Word	.docx	20MB	Text extraction via mammoth.js
Markdown	.md	20MB	Read directly as text
Text	.txt	20MB	Read directly as text

Starting Processing

Once content collection/upload is complete, click “Start Processing.”

Duplicate Processing Prevention: If “Start Processing” is clicked for a concierge already being processed, an “Already processing” (HTTP 409) error is displayed.

13.4 Step 2: Processing

The automated pipeline runs. No user action needed; everything proceeds automatically.

Processing Steps Detail

Step	Description	Estimated Time
1. Extracting concepts	Split text into chunks and extract concept elements with AI	Depends on data volume (5-10 sec per chunk)
2. Uploading data	Send extraction results to ConceptMiner engine	A few seconds
3. Generating embeddings	Text vectorization (OpenAI text-embedding-3-small)	10-30 seconds
4. Dimension reduction	Compress to 6 dimensions with UMAP	5-15 seconds
5. Building model	GNG network learning	10-30 seconds
6. Clustering	Group by theme with Ward clustering	5-10 seconds
7. Saving	Persist model data	A few seconds

Progress Display:

Each step’s name and progress are shown
Concept extraction step shows chunk-level progress like “X/Y chunks processed”
Auto-polled every 3 seconds

Processing Complete:

All steps complete, automatically advances to the next step (Customize)

Processing Failed:

Error message is displayed
“Retry” button returns to source selection

Credit Cost:

Concept extraction: 2 credits per chunk
Embedding generation: 10 credits
Model building: 20 credits
Total: 30 + chunk count × 2 credits

13.5 Step 3: Customization

Customize the widget’s appearance and behavior.

Greeting Message

Set the message displayed when the widget opens.

Operation: Enter message in text input field

Examples:

“Hello! How can I help you?”
“Ask me anything about our products.”
“Need help? Feel free to ask!”

Theme Color

Set the widget’s accent color.

Operation: Select with color picker or enter HEX code (e.g., #2563eb)

This color is applied to the widget header, send button, links, etc.

Proactive Messages

Set messages the widget automatically displays before visitors ask questions. Useful for broadcasting news and announcements.

Operation:

Enter message in text input field
Click “+” to add
Messages appear in a list
Remove unwanted messages with trash icon

Examples:

“We just launched our new product XYZ!”
“March-only campaign running now.”
“Business hours: Weekdays 9:00-18:00.”

Preview

Widget appearance preview is shown at the bottom of the screen. Configured colors, greetings, and proactive messages are reflected in real-time.

When customization is complete, click “Save and Continue to Publish Settings.”

13.6 Step 4: Publishing

Embedding Code

HTML code like the following is displayed:

<script
  src="https://your-domain/api/concierge/widget.js"
  data-api-key="xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
  data-color="#2563eb"
  async
></script>

Operation: Click “Copy” to copy the code to clipboard

Installing on Your Website

Open your website’s HTML
Paste the copied