Setting Up Knowledge Bases

Learn how to create and manage knowledge bases that organize your documents and make them available to your agents.

Creating a Knowledge Base

1. Navigate to Admin Console → Knowledge Base
2. Click "Add Knowledge Base"
3. Enter a descriptive name (e.g., "company-policies", "technical-docs")
4. Click "Create"

Organization Examples

Structure by department or function:

• hr-policies - Employee handbook, benefits
• engineering-docs - Technical documentation, APIs
• customer-support - FAQs, troubleshooting guides

Uploading Documents

Supported Formats

• PDF (.pdf), DOCX (.docx), XLSX (.xlsx), Text (.txt), Markdown (.md), CSV (.csv), SRT (.srt), RTF (.rtf), and images (.jpg, .jpeg, .png, .gif, .webp)
• Maximum size: 50MB per file

Upload Methods

1. Expand your knowledge base (click the header)
2. Drag & drop files onto the upload area, or click "Select Files"
3. Monitor progress - files show as Processing → Success/Error

Adding Text Manually

For quick content without files:

1. Click "Add Text" in the expanded knowledge base
2. Enter content in the text area
3. Click "Upload"

Managing Documents

Search and Organization

• Search bar - Find documents by filename or author
• Document table - Shows name, author, status, date
• Model warning badge - Shows when a resource was embedded with a different model than the current setting, or when its chunks use mixed embedding models
• Delete documents - Click trash icon in Actions column
• Delete knowledge base - Click trash icon in header (⚠️ permanent)

Embedding Model Notes

• Knowledge-base content is embedded into halfvec(1536) vectors.
• The default embedding model is text-embedding-3-small.
• Supported knowledge-base models are text-embedding-3-small, gemini-embedding-001, and gemini-embedding-2.
• Gemini embedding models are stored at 1536 dimensions.
• text-embedding-3-large is not available in the Knowledge Base embedding-model selector.

Changing the embedding model does not rewrite existing resources automatically. Existing chunks keep their stored embedding_model; search generates query vectors for the stored models it finds and only compares chunks against vectors from the same model. Regenerate a resource or knowledge base to clear model mismatch or mixed-model warnings.

Best Practices

• Group related content - Keep similar documents together
• Use descriptive names - employee-benefits-2024.pdf not document1.pdf
• Keep content current - Replace outdated versions regularly
• Plan agent access - Match knowledge bases to agent roles

Troubleshooting

Upload failures:

• Check file size (under 50MB) and format (PDF, DOCX, XLSX, TXT, MD, CSV, SRT, RTF, JPG, JPEG, PNG, GIF, WEBP)
• Verify file isn't corrupted by opening locally

Content not searchable:

• Ensure file shows "Success" status
• Verify agent has searchKnowledgeBase tool enabled
• Confirm knowledge base is assigned to agent
• If a model warning is present, search can still work, but regenerate the resource to align it with the current embedding model

Connecting to Agents

1. Go to Agents → Select your agent → Knowledge Base section
2. Enable specific knowledge bases or Enable all
3. Save changes

Test by asking your agent questions about your uploaded content!