Reference Document Analysis - User Guide
This guide explains how to upload reference material (PDF) to a project, turn it into a form the AI can read, and then ask the chatbot about its content. It covers the upload options (processing mode and model tier), the processing steps and document statuses, how to query the chatbot, and the cost.
Last Updated: 2026-02-21
1. Introduction
TFLOW's Reference Document Analysis feature uses AI to analyze uploaded PDF documents and convert them into a Knowledge Graph. The converted documents become the foundation that lets the AI chatbot accurately understand the document content and answer questions about it.
When you upload a document, you can select a Processing Mode and a Model Tier.
Per-Project Document Isolation
Documents uploaded to a project are searched only within that project. A document uploaded in Project A will not appear in Project B's search results, so each project maintains its own independent knowledge base.
2. Uploading Documents
2.1 Accessing the Upload Screen
- On the project detail page, select the Reference Documents tab
- Click the Upload button to open the upload modal
- Drag and drop a PDF file, or click to select one
2.2 Selecting Processing Options
Two options are displayed below the upload area.
Processing Mode
| Mode | Description | Recommended For |
|---|---|---|
| Realtime | Processing starts immediately and completes quickly | Small documents (10 pages or fewer), urgent cases |
| Batch (50% OFF) | Batch processing reduces the cost by 50% | Large documents (20+ pages), when cost savings matter |
When you select Batch mode, document analysis runs in the background and you receive a notification when it completes. Most jobs finish within 10 minutes.
Model Tier
| Tier | Description | Recommended For |
|---|---|---|
| Premium (Opus) | Uses the highest-quality AI model | Complex technical documents, when high accuracy is required |
| Standard (Sonnet) | The default model, balancing quality and cost | General documents (default) |
| Basic (Haiku) | Uses the most economical model | Simple documents, when cost is the top priority |
2.3 Running the Upload
After selecting your options, click the Upload button to start document analysis. Once the document status changes from analysing to indexed, you can query that document in the AI chatbot. Batch mode is processed in the background, so you can close the screen and you will be notified when it completes.
3. Processing Flow
Realtime Mode
- Analysis starts immediately upon upload
- The document status changes from
analysingtoindexed - Once complete, you can immediately query the document content in the AI chatbot
Batch Mode
- After upload, a batch job is registered
- Processing runs step by step in the background:
- Document image conversion and OCR
- Markdown conversion
- Section splitting and chunking
- Knowledge triple extraction
- Knowledge Graph loading
- Section summary generation
- You receive a notification when it completes
- You can check the status in the document list
Document Status
| Status | Meaning |
|---|---|
| Uploading | File is being uploaded |
| Waiting | Waiting to be processed |
| Analysing | Analysis in progress |
| Indexed | Analysis complete, available in the AI chatbot |
| Failed | Analysis failed (contact your administrator) |
4. Querying Documents with the AI Chatbot
Once a document reaches the Indexed status, you can ask questions about its content in the CYGNUS AI chatbot.
4.1 Using the Chatbot
- Enter your question in the chat area at the bottom of the Reference Documents tab
- The AI searches the project's indexed documents and answers
- The answer includes reference information for the source documents it found
4.2 Search Modes
| Mode | Description |
|---|---|
| Auto | The AI automatically selects the optimal search method based on the question type (default) |
| Hybrid | A combination of vector and full-text search |
| Vector | Semantic similarity search |
| Fulltext | Keyword-based full-text search |
| Section | Search at the document-section level |
4.3 Document Filtering
When you want to query only specific documents, you can use the document filter to narrow the search scope. If no filter is set, the search covers all indexed documents in the current project.
5. Cost Guide
Document analysis consumes AI model usage tokens. The cost varies by processing mode and model tier.
Cost Comparison (based on a 100-page document)
| Combination | Relative Cost |
|---|---|
| Realtime + Premium | ★★★★★ (highest) |
| Realtime + Standard | ★★★ |
| Realtime + Basic | ★★ |
| Batch + Standard | ★★ (50% savings vs. Realtime) |
| Batch + Basic | ★ (most economical) |
You can check your remaining tokens in the token gauge on the top navigation bar.
6. FAQ
Processing time and model choice
Q: I uploaded in Batch mode. When will it finish?
Most jobs finish within 10 minutes. The time can vary depending on document size and server load, and may take up to 24 hours. You receive a notification when it completes.
Q: Can I do other work while analysis is running?
Yes. Batch mode runs entirely in the background, so you are free to do other work. In Realtime mode, analysis also continues even if you navigate to another page.
Q: Which model tier should I choose?
For documents where accuracy matters, such as technical specifications or contracts, we recommend Standard or higher. For simple reference materials, Basic is sufficient.
Q: My analysis failed. What should I do?
Delete the document from the document list and upload it again, or contact your administrator. Note that tokens used are still deducted even when analysis fails.
Uploads, tokens, and project isolation
Q: Can I upload multiple documents at once?
Yes. You can drag and drop or select multiple files at the same time. Each document is processed independently.
Q: What happens if I run out of tokens?
If your balance is 0 or below, analysis will not start. Ask your administrator to top up your tokens.
Q: Can I see documents uploaded in other projects?
No. Each project's documents are completely isolated. A document uploaded in Project A is searchable only in Project A's chatbot and cannot be accessed from Project B.
Q: I want to use the same file across multiple projects.
You must upload it separately in each project. Even for an identical file, an independent Knowledge Graph is created for each project.
7. Related Documents
- CYGNUS Batch Processing Administrator Guide (internal document — not published on this site)