5.4. Complete Guide to Lyra Interactive Capabilities

Lyra is not just a Q&A chatbot. It is an interactive AI assistant that can perform actual work on your behalf — filling out forms, analyzing files, exporting data, navigating pages, and more. This guide walks through every capability Lyra offers, with examples.

Capabilities at a Glance

1. Form Autofill — "Fill It Out for Me"

Lyra can automatically populate the empty form fields on the current screen. The AI analyzes project information, attached documents, and prior data to suggest appropriate values.

How It Works

1. Lyra suggests values to enter into the form fields.
2. Apply All / Decline options appear on the screen.
3. Clicking Apply All automatically enters the suggested values into the form.
4. After reviewing the entered values, you must click the Save/Submit button yourself.

💡 Lyra only fills out the form; it does not automatically submit it. The final save is always performed by you.

Try Saying This

• 💬 "Fill out this form for me"
• 💬 "Auto-fill the project information"
• 💬 "Complete this form based on the CI"
• 💬 "Fill in all the blanks"

Example Pages Where This Works

FAQ

• Q: "I clicked Apply All, but only some fields were filled."
  • A: Disabled or read-only fields cannot be autofilled. You must enter those fields manually.
• Q: "A filled value is incorrect."
  • A: You can press Decline or edit the value manually. Lyra's suggestions are for reference only, and final confirmation is up to you.

2. File Analysis — "Analyze This File"

When you attach a file in the chat window, Lyra analyzes it in various ways. It supports multiple formats including PDF, images, and Excel.

7 Types of Analysis

Try Saying This

• 💬 "Summarize this PDF"
• 💬 "What is the total on this Invoice?"
• 💬 "Compare the CI and PL to check whether the quantities match"
• 💬 "Extract the HS Code from this document"
• 💬 "Find the red-pen corrections"

⚠️ To request file analysis, you must first attach a file in the chat window. If you request analysis without an attachment, Lyra has nothing to analyze.

FAQ

• Q: "I uploaded a file, but it won't analyze."
  • A: Check whether the file is a supported format (PDF, image, Excel, Word, etc.). Very large files may take time to process.
• Q: "How do I compare two files?"
  • A: Attach both files, then ask "Compare these two files."

3. Document Registration — "Register It to the Project"

You can register a file attached in the chat window as an official document of the current project. Registered documents are managed in the project's Documents tab.

How It Works

1. Attach a trade document (CI, PL, BL, etc.) in the chat window.
2. Ask Lyra to register it.
3. Specifying the document type speeds up AI analysis.
4. The system registers the document to the project and starts AI analysis.

Supported Document Types

Commercial Invoice, Packing List, Bill of Lading, Airway Bill, Letter of Credit, Purchase Order, Proforma Invoice, Certificate of Origin, Quality Certificate, Inspection Certificate, and more.

Try Saying This

• 💬 "Register this file to the project"
• 💬 "Save this CI as a project document"
• 💬 "Register and analyze this B/L"

💡 If you also state the document type (e.g., "Register this CI"), you can reduce the time the AI spends automatically determining the document type.

4. Page Navigation — "Take Me to ~"

Just tell Lyra and you can instantly navigate to the page you want. No need to find and click through menus — request it in natural language.

Navigation Types

Try Saying This

• 💬 "Go to the project list"
• 💬 "Open the email composition screen"
• 💬 "Take me to the payment screen"
• 💬 "Go back"
• 💬 "Return to the previous screen"

💡 Which pages you can navigate to depends on your current access permissions. Navigation to pages you don't have permission for may be restricted.

5. Data Lookup — "Tell Me the Count of ~"

Lyra converts natural-language questions into database queries to look up project, trade, document, and business partner information.

Data You Can Look Up

Try Saying This

• 💬 "Show me the status of projects completed this month"
• 💬 "Show me the list of unprocessed documents"
• 💬 "Look up the list of USD-currency projects"
• 💬 "Does partner ABC have any cases in progress?"

💡 When there are many results, the top 10 are shown by default. If you need more results, ask "Show me all of them."

6. Export — "Export to Excel"

You can export lookup results, table data, analysis reports, and more in the format you want. Four formats are supported — Excel, CSV, JSON, and PDF — so pick the one that matches what you need it for: spreadsheet work, data exchange, programmatic use, or a printable report.

Supported Formats

Try Saying This

• 💬 "Export this data to Excel"
• 💬 "Download the project list as CSV"
• 💬 "Make a PDF report from the analysis results"

7. Screen Capture and Reading — "Look at the Current Screen"

Lyra can read the entire content of the current screen, or analyze charts/images through a visual capture.

Two Methods

Try Saying This

• 💬 "Read out all the content on the current screen"
• 💬 "Explain this chart"
• 💬 "Analyze the data shown on the screen"

💡 Lyra automatically recognizes the key information on the screen even under normal use. However, when content is truncated or visual analysis is needed, explicitly requesting this feature yields more accurate answers.

8. Import Readiness Review (IRR) — "Are We Ready for Customs Clearance?"

This comprehensively reviews a project's trade documents and import requirements to provide a report on customs-clearance readiness. The 6-step automated pipeline is visualized in a multi-stage modal, letting you check each step's progress and intermediate results in real time. ✨ 2026-04 update

6-Step Review Process (Multi-Stage)

1. Data collection: Collect project information, document list, and trade data
2. Document cross-verification: Check whether quantities, amounts, and item names match across the LC, CI, PL, and BL
3. Import requirement search: Look up required documents, certifications, and regulations by destination country and item (with source citations)
4. Issue classification: Classify discovered problems into 3 levels (Blocker/Warning/Confirmation Required)
5. Verdict determination: Determine the final import readiness status
6. Report generation: Save the comprehensive report to the database

💡 Reviewing intermediate results: A result block is added to the modal as each step completes, so you can preview some results even before the full analysis finishes. Even if you leave the page mid-analysis and come back, it is restored to the REVIEWING state so your progress is not lost.

💡 Source traceability: Each requirement/regulation match result is linked to a source_registry-based source (government site, official document), so you can immediately review the basis for the verdict.

Verdict Results

Issue Types

Try Saying This

• 💬 "Is this project ready for customs clearance?"
• 💬 "Review the import readiness status"
• 💬 "Check whether it meets Dubai's import requirements"
• 💬 "Cross-verify the documents"

⚠️ HS Code recommendations are reference guidance only and have no legal effect. For final HS Code determination, confirm with a customs broker or the competent customs office.

9. Python Analysis — "Analyze the Data"

Lyra runs Python code in a secure sandbox environment to perform data analysis, statistical calculations, and visualization.

Available Libraries

pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, openpyxl, tabulate

Usage Examples

Try Saying This

• 💬 "Analyze this data and make a chart"
• 💬 "Draw a graph of quarterly sales trends"
• 💬 "Do a statistical analysis of these numbers"
• 💬 "Visualize the lookup results as a chart"

💡 For security, external network access and system command execution are restricted. It is an environment optimized for data analysis and visualization.

10. Knowledge Search — "Tell Me About UCP 600 Article 14"

This searches TFlow's internal knowledge base for system usage, trade regulations, and practical knowledge. It provides verified, accurate information rather than external web search.

Search Modes

Try Saying This

• 💬 "What can I do on this screen?"
• 💬 "Explain what this field is"
• 💬 "Tell me cases related to UCP 600 Article 14"
• 💬 "What is a Commercial Invoice?"
• 💬 "Give me the import checklist for tomato sauce"

💡 Lyra automatically recognizes the screen you are currently viewing and prioritizes help relevant to that page.

11. Web Search — "Show Me the Latest Exchange Rate"

When you need up-to-date information that is not in the internal knowledge base, Lyra performs a web search. It is used for information that changes in real time such as exchange rates and fuel prices, recent regulation changes, and reading the content of a URL you attach — and the results always come with their source URLs.

When to Use

Try Saying This

• 💬 "Tell me today's USD/KRW exchange rate"
• 💬 "Are there any recent changes to Dubai's import regulations?"
• 💬 "Read the content of this link" (with a URL attached)

⚠️ Web search results come from external sources, so they may be less accurate than the internal knowledge base. Verify the source of important information. Lyra always provides the source URL alongside its answer.

12. HS Code AI Classification — "Give Me the HS Code for This Product"

Enter a product description, and the AI automatically recommends a suitable HS Code through a 4-layer pipeline. It supports everything from the international 6-digit HS Code to country-specific extended codes such as Korea (HSK 10-digit) and UAE/GCC (8-digit).

AI Classification Pipeline

1. Attribute extraction (L1): The AI extracts key attributes such as material, use, form, and processing state from the product description
2. Graph traversal (L2): Traverses the HS Code hierarchy in the Neo4j knowledge graph to derive candidate codes
3. Case search (L3): Matches similar past classification cases (advance ruling precedents) via vector search
4. Final classification (L4): The AI synthesizes all information to recommend the optimal HS Code and present its rationale

Supported Countries

Try Saying This

• 💬 "Give me the HS Code for a semiconductor wafer"
• 💬 "Classify the Korean HSK code for a lithium-ion battery"
• 💬 "What is the HS Code for tomato sauce when importing to Dubai?"
• 💬 "Check the HS Codes for the products in this CI"

HS Code Search Screen

In addition to the Lyra chat, HS Code search is also available directly on a dedicated search screen.

• Select the HS Code menu in the left menu, or ask Lyra to "Open the HS Code search screen."
• Enter a product description, and the AI shows a list of recommended HS Codes in card form.
• Each card displays the code, item name, tariff rate, and classification rationale.

⚠️ AI HS Code recommendations are reference guidance only and have no legal effect. For final HS Code determination, confirm through a customs broker or an advance ruling from the competent customs office.

FAQ

• Q: "Multiple HS Codes came up — which one should I use?"
  • A: The AI sorts them by confidence. Review the top recommendation first, but if there are similar candidates, choose the code that best fits your product's characteristics. If you're unsure, we recommend confirming with a customs broker.
• Q: "No code that fits my company's product came up."
  • A: Try entering a more specific product description. For example: "plastic bottle" → "transparent 500ml PET beverage bottle, cap not included." Including material, use, size, and processing state improves accuracy.

13. Trade Term Search — "What Is FOB?"

This searches TFlow's built-in trade term dictionary of 841+ Korean/English trade terms. It provides not only the term's definition but also related terms and practical usage.

Search Methods

• Full-text search: Searches across the term name, abbreviation, and description text
• Semantic search: Also recommends semantically similar terms
• Related term exploration: Automatically displays connected terms based on the knowledge graph

Try Saying This

• 💬 "What is FOB?"
• 💬 "Tell me the difference between CIF and CFR"
• 💬 "What does L/C at sight mean?"
• 💬 "What is a surrender B/L in trade?"
• 💬 "Tell me the types of Incoterms"

Glossary Screen

In addition to the Lyra chat, you can explore terms by category on a dedicated term search screen.

• Select the Glossary menu in the left menu, or ask Lyra to "Open the glossary."
• Explore by category: Incoterms, payment methods, transport, insurance, customs, documents, and more
• Real-time autocomplete and result display as you type a search term

💡 When you ask Lyra a single-term question like "What is FOB?", it looks up the definition in the glossary (search_glossary). When you need term comparisons, practical context, or detailed explanations, knowledge search (trade mode) is used.

Term Search vs. Knowledge Search — What's the Difference?

FAQ

• Q: "I only know the English abbreviation — can I still search?"
  • A: Yes. You can search by abbreviation (e.g., B/L, AWB, T/T), the full English name, and the Korean name.
• Q: "The term explanation is too brief."
  • A: The default explanation on a term card is a summary. If you ask Lyra a follow-up like "How is this term used in practice?", you'll get a more detailed answer with practical examples.

14. Automatic BL Correction — "Correct This BL"

This automatically compares a B/L (Bill of Lading) registered to a project against reference documents (LC, PL, CI), identifies items that need correction, and generates correction suggestions.

How It Works

1. The AI compares the B/L content against the project's reference document chain (LC → PL → CI).
2. It automatically detects mismatched or missing items.
3. For each item, it generates a correction suggestion (REPLACE/ADD/DELETE).
4. You can visualize the results as a red-pen markup PDF or review them as a correction list.

Correction Suggestion Types

Try Saying This

• 💬 "Correct the BL for this project"
• 💬 "Compare the BL and LC content and tell me what's wrong"
• 💬 "Check the Bill of Lading for errors"

Reviewing Results

• Correction list view: Correction suggestions are organized into a table. You can review each item's current value, correct value, and reason for correction.
• PDF markup view: Correction locations are visually marked with red-pen annotations over the original B/L.
• Excel export: You can download the correction details as an Excel file.

💡 For BL Correction analysis, only users with the Seller role can edit corrections; Buyers can only view the results.

FAQ

• Q: "There's no BL yet — can I still run correction analysis?"
  • A: No. A B/L document must be registered to the project. You also need at least one reference document to compare against (one or more of LC, CI, PL).
• Q: "Can I apply the correction results directly to the B/L?"
  • A: The correction suggestions are for reference; the actual B/L revision must be requested from the shipping line. Export the correction list to Excel and send an amendment request to the shipping line.

15. LC Risk Analysis (LEO) — "Analyze the Risk of This LC"

LEO (Letter of Credit Evaluation through Ontology-driven Inference) is an engine in which the AI automatically analyzes the risk of a Letter of Credit (LC). Based on UCP 600 rules, ISBP standards, and past dispute cases, it proactively detects potential problems in an LC.

Analysis Pipeline

LEO validates an LC from multiple angles through a 4+1 layer pipeline:

Risk Report

The analysis results are provided as a report containing the following items:

• Discrepancy list: Detailed descriptions for each discovered mismatch/risk item
• Risk level: HIGH / MEDIUM / LOW rating
• Source documents: The UCP article, ISBP rule, and similar dispute cases for each risk
• Confidence score: The final confidence after EVA verification

Try Saying This

• 💬 "Analyze the LC risk for this project"
• 💬 "Show me the LC verification results"
• 💬 "Check whether there are any risky terms in the LC"

Requirements

• An LC document must be registered to the project
• The LC document must be in a structured state through analysis (the CYGNUS pipeline)

💡 LEO analysis can run automatically upon document registration, or be requested manually. You can also retrieve the results from external tools via MCP using get_leo_report.

FAQ

• Q: "The LC is still being analyzed — can I see the LEO results?"
  • A: LEO analysis is only possible after the LC document has finished parsing and analysis through the CYGNUS pipeline. You can check the analysis status in the document list.
• Q: "Are there cases where an item LEO flagged as risky is actually fine?"
  • A: The EVA (L5) layer filters out false positives, but the final judgment must be made by an expert. Use the LEO results as a review aid.

16. Coach DOM Interaction — "Where Is This Button Here?" ✨ NEW

In AI Coach tutorial mode, Lyra uses 3 dedicated tools that let it directly query and manipulate the screen DOM. As a result, instead of merely saying "click here" in words, it can point to actual screen elements or click/type on your behalf.

3 Coach-Only Tools

⚠️ act_element can only be used in Coach mode (tutorial), and a confirmation dialog appears before execution so it always requires your approval. During normal Lyra conversation it is not invoked outside of tutorials.

Try Saying This

• 💬 "Where is the Save button here?" → get_page_elements + highlight
• 💬 "Turn on the Publish toggle" → act_element (executed after confirmation)
• 💬 "What should I enter in this field?" → explains the label, placeholder, and required status via inspect_element

FAQ

• Q: Does it also click buttons for me when not in Coach mode?
  • A: No. In normal conversation, Lyra only guides (highlights), and you must click yourself. Coach tools are enabled only within tutorial courses.
• Q: How does it choose when there are multiple buttons?
  • A: Lyra combines the label, position, and tooltip to propose the most likely candidate, and if it's ambiguous, it asks you back, "Do you mean A or B?"
• Q: How do I prevent act_element from accidentally pressing an important button like Save?
  • A: Every execution requires your confirmation, and irreversible actions (delete, payment execution) are classified in the Coach tool block list.

17. Composite Workflows — Combining Multiple Capabilities

Lyra's true value is that it can combine multiple capabilities to handle complex tasks in one go.

Representative Workflows

Creating a CI based on a PO

1. Attach a PO (Purchase Order) file for analysis
2. Navigate to the CI (Commercial Invoice) creation screen
3. Autofill the CI form based on the PO analysis data

Export after document analysis

1. Analyze the attached file (data extraction)
2. Export the extracted data to Excel

Comprehensive import readiness review

1. Collect all project documents
2. Cross-verify the documents
3. Search import requirements
4. Generate a comprehensive verdict report

HS Code check + import requirement research

1. Classify the HS Code from the product description via AI
2. Automatically search the destination country's import requirements using that HS Code
3. Provide a checklist of required documents and certifications

BL correction + drafting a shipping-line request email

1. Run correction analysis between the B/L and reference documents
2. Export the correction details to Excel
3. Draft an amendment request email to send to the shipping line

Try Saying This

• 💬 "Analyze this PO and fill in the CI form"
• 💬 "Analyze the attached file and export the results to Excel"
• 💬 "Review all the project documents and tell me whether we can clear customs"
• 💬 "Find this product's HS Code and check Dubai's import requirements"
• 💬 "Correct the BL and draft an email to send to the shipping line"

💡 For composite requests, Lyra automatically executes the necessary steps in order. If confirmation is needed along the way, Lyra will ask.

Frequently Asked Questions (FAQ)

• Q: "What's the criterion for what Lyra can and cannot do?"
  • A: Lyra can perform tasks such as information lookup, analysis, form-fill suggestions, and data export. However, hard-to-reverse actions such as executing payments, concluding contracts, or changing data in external systems require user approval or are not supported.
• Q: "Can I ask Lyra to do several things at once?"
  • A: Yes. If you make a composite request like "Analyze this file and register it to the project," Lyra will handle it in order.
• Q: "Can I make requests in English too?"
  • A: Yes. Lyra supports both Korean and English. It responds naturally in the language you use.
• Q: "A value Lyra recommended seems wrong."
  • A: Lyra's suggestions are AI-based, so they are not always 100% accurate. Select Decline or edit the value manually. Always verify important data before use.
• Q: "I want to cancel a task I asked Lyra to do."
  • A: For form autofill, pressing Decline prevents it from being applied. If it has already been applied and you haven't saved/submitted yet, you can refresh the page to return to the original state.