Knowledge Base (RAG)
To understand the Knowledge Base, it helps to understand a concept called RAG (Retrieval-Augmented Generation).
Think of a standard AI model (like ChatGPT) as a genius student taking a closed-book exam. It relies entirely on its memory of general facts it learned during training. While smart, it cannot answer questions about your specific private data—such as your company's latest presentation schedule or internal financial reports—because it never studied them.
The Knowledge Base changes the test to an "Open Book" exam.
By connecting a RAG node, you hand the agent a textbook containing your specific files. Before answering, the agent:
- Searches through your uploaded documents for relevant context
- Reads the specific details found
- Constructs an answer based on your facts, avoiding hallucination
Visual Configuration
In VibeAgent, memory is not a hidden setting; it is a physical component of your graph. You configure it visually by connecting a Knowledgebase Node to your agent.
Add the RAG Node
Locate the Knowledgebase node in your left-hand library and drag it onto the canvas. This node acts as the container for your documents.
Connect to Agent
Draw a connection line from your Agent Node to the Knowledgebase Node.
- Visual Cue – This explicitly tells the system, "This specific agent has permission to read these specific documents"
Upload Data
Click on the Knowledgebase Node itself to open the upload modal. You have three ways to add data:
- Select Files – Upload PDFs, Word docs, or text files directly from your computer
- Add URL – Paste a specific web link; the system will scrape the text and add it to the knowledge base
- Google Drive – Connect your Drive account to import documents directly from the cloud
File Processing States
After uploading, each file displays a status badge:
| Status | Badge Color | Description |
|---|---|---|
| Processing | Orange | File is being parsed and indexed. Large files may take longer. |
| Ready | Green | File successfully processed and available for RAG queries. |
| Failed | Red | Upload or processing error. Click the file to view error details. |
Refresh Button: If files remain in "Processing" state for an extended period, click the refresh icon (circular arrow) to update the status.
If a file fails, common causes include:
- File size exceeds 100MB (check the "Supported: PDF, DOCX, CSV, Text, JSON (Max 100MB)" header)
- Corrupted or password-protected files
- API quota exceeded (see error message for details)
Delete the failed file and re-upload after resolving the issue.
Wait for all files to show the green Ready badge before clicking Done.
Schedule Periodic Re-embedding (URLs Only)
Keep your knowledge base up to date by scheduling automatic re-embedding for URLs. When website content changes, the agent can automatically refresh its indexed data.
How it works:
- Each URL has a schedule dropdown (displayed next to the URL entry)
- Select a frequency: No schedule, Every 1 day, Every 3 days, Every 7 days, Every 14 days, or Every 30 days
- The system automatically re-scrapes and re-embeds the URL content at the selected interval
- No notifications or indicators when re-embedding occurs—it happens silently in the background
Important Notes:
- ✅ Applies only to URLs (website links)
- ❌ Does not apply to uploaded files (PDF, DOCX, CSV, TXT)
- Each URL has its own independent schedule
- Set different schedules for different URLs based on how frequently their content updates
Best For:
- News websites that publish daily updates
- Documentation pages that change regularly
- Product catalogs with frequent inventory changes
- Company pages with evolving information
Use "Every 1 day" for frequently updated content (news, blogs), "Every 7 days" for documentation, and "Every 30 days" for relatively static pages.
Supported File Types
VibeAgent supports files up to 100MB in size. The system automatically chunks and indexes text from these formats:
| Format | Extension | Best For |
|---|---|---|
.pdf | Official reports, slide decks, and contracts | |
| Word | .docx | Editable policies and drafts |
| Structured Data | .csv, .json | Product catalogs, contact lists, or raw data exports |
| Plain Text | .txt | Simple notes and developer logs |
Testing & Verification
To ensure the connection is working, use the Run button to simulate a user query. Ask a question that is specific only to your uploaded document.
Test Question: "When is the presentation scheduled?" (assuming you uploaded a schedule PDF)
Success Criteria: The agent will cite the specific time and date found in your document rather than giving a generic answer.