Skip to main content
title
Prerequisites: Have your business content ready (FAQs, product docs, policies). See Best Practices first.
Augment your voice agents with specialized knowledge through AI-powered retrieval augmented generation (RAG), ensuring agents have access to your business information during every call.

​
What is Genius?

Genius databases serve as intelligent knowledge repositories that inform your voice agents about your business, services, and any domain-specific information. Using retrieval augmented generation technology, Genius transforms your source materials into optimized question-and-answer formats that agents can easily understand and reference during calls.

​
How Genius Works

  1. Content Ingestion: You upload various file types or provide URLs containing your business information
  2. AI Processing: The system analyzes and transforms content into structured Q&A format
  3. Agent Integration: Genius databases are connected to specific agents through Automations or Agent Builder
  4. Call Enhancement: During calls, agents access relevant information to provide accurate, informed responses

​
Supported Content Types

Genius accepts multiple content formats to accommodate different information sources:

​
File Types

  • Text files: Plain text documents with business information
  • PDF documents: Reports, manuals, guides, and formatted documents
  • Audio files: Recorded training materials or information sessions
  • CSV files: Structured data (see best practices below for optimal usage)

​
Web Content

  • URLs: Website pages, documentation sites, and online resources
  • Static content: Information is fetched once during setup and not automatically updated

​
Best Practices for Content Types

​
Recommended: Unstructured Data

Why Unstructured Works Better: RAG technology excels at processing natural language content and converting it into contextual knowledge that agents can understand intuitively. Ideal Content Formats:
  • Narrative descriptions of business processes
  • Customer service scripts and guidelines
  • Product documentation written in conversational style
  • FAQ documents with natural language questions and answers

​
Use with Caution: Structured Data

CSV and Spreadsheet Limitations: While CSV files are supported, they can lead to higher rates of hallucinations or inaccurate responses due to the structured nature conflicting with RAG processing.
Common Mistake: Uploading large spreadsheets with complex data relationships. RAG works best with natural language. Convert structured data into Q&A format instead: “Q: What’s the price for Product A? A: $299.”
When to Use CSV:
  • Simple lookup tables with clear key-value relationships
  • Small datasets with straightforward information
  • Data that supplements rather than replaces unstructured content
CSV Best Practices:
  • Keep data simple and avoid complex relationships
  • Include descriptive headers that provide context
  • Combine with unstructured explanations when possible

Content type comparison

​
Getting Started

​
Step 1: Create a Genius Database

  1. Navigate to Genius in the sidebar
  2. Click “Create New Genius” button
  3. Enter a descriptive database name
  4. Save to create your new knowledge base
The new database appears in the sidebar and is ready for content addition.

Building your Genius

​
Step 2: Add Content Sources

  1. Select your newly created database from the sidebar
  2. Choose content type (Text, Audio, PDF, CSV, or URL)
  3. Provide a descriptive name for the source
  4. Upload file or enter URL as appropriate
  5. Save to begin processing

​
Processing Time

Content processing typically takes 2-10 minutes depending on:
  • File size and complexity
  • Content type (text processes faster than audio)
  • Current system load
A progress indicator shows processing status in the Genius source table.
Content processing interface

Content processing interface

​
Content Management

​
Viewing Processed Content

After processing completes:
  1. Click the three-dot menu next to any source in the table
  2. Select “View” to see processed content
  3. Review how your content was transformed into Q&A format
The processed format shows how agents will understand and access your information during calls.

​
Managing Sources

Edit Sources: Use the three-dot menu to modify source names or settings Delete Sources: Remove outdated or incorrect content sources Add Multiple Sources: Build comprehensive databases with diverse content types

​
Database Management

Database Settings: Click the gear icon to modify database name or settings Delete Database: Permanently remove entire databases and all contained sources Duplicate Content: The same source can be added to multiple databases if needed

​
Agent Integration

​
Connecting to Agents

Genius databases must be explicitly connected to agents through: Automations Interface:
  • Select Genius database within automation workflows
  • Configure how knowledge is accessed during automated processes
Agent Builder Interface:
  • Choose Genius database for specific agent configurations
  • Set up when and how knowledge is retrieved during calls

​
Multiple Database Support

  • Different agents can use different Genius databases
  • Single agents can access multiple databases if configured
  • Scope selection happens during automation or agent setup
Agent integration options

Agent integration options

​
Content Optimization

​
Writing for RAG Processing

Use Natural Language: Write content as if explaining to a knowledgeable colleague Include Context: Provide background information that helps agents understand when to use specific information Ask Questions: Include common customer questions and comprehensive answers Be Specific: Avoid vague statements that could lead to ambiguous agent responses

​
Example Optimization

Less Effective: 
Product A: $50
Product B: $75
Product C: $100
More Effective: 
Our product pricing is structured to meet different customer needs:

Product A is our entry-level solution at $50, ideal for small businesses just getting started with our service. It includes basic features and email support.

Product B at $75 represents our most popular package, offering advanced features, phone support, and integration capabilities that most growing businesses require.

Product C at $100 is our enterprise solution with all premium features, dedicated account management, and custom integration support for large organizations.

​
URL Content Considerations

​
Static Content Only

Important Limitation: URLs are fetched once during setup and are not automatically refreshed or updated. Best Practices for URLs:
  • Use stable, long-term content sources
  • Avoid frequently changing pages
  • Consider downloading and uploading as files for content you control
  • Plan to manually update URL sources when content changes significantly

​
URL Content Quality

Choose High-Quality Sources: Select comprehensive, well-written pages that provide complete information Avoid Dynamic Content: Pages with frequently changing information may become outdated Test Accessibility: Ensure URLs are publicly accessible and not behind authentication

​
Safety and Accuracy

​
Source Citation

Genius provides both approaches for information delivery:
  • Behind-the-scenes processing: Enhanced agent knowledge without explicit citation
  • Source attribution: Agents can reference specific sources when appropriate

​
Content Grounding

The system is designed to ground agent responses in provided source material, reducing hallucinations and ensuring accuracy. However, content quality directly impacts response quality.

​
Quality Control

Review Processed Content: Always check how your sources were transformed Test Agent Responses: Verify agents provide accurate information from your sources Update Regularly: Remove outdated sources and add current information Monitor Performance: Track how well agents utilize Genius knowledge in real calls

​
Expected Results

After successful Genius setup and integration: Enhanced Agent Knowledge:
  • Agents provide accurate, company-specific information during calls
  • Consistent responses across all agents using the same database
  • Reduced need for agent training on factual information
Improved Call Quality:
  • Faster resolution of customer questions
  • More detailed and accurate responses
  • Professional handling of complex or specialized inquiries
Operational Benefits:
  • Centralized knowledge management across all agents
  • Easy updates to business information affecting multiple agents
  • Scalable knowledge distribution for growing teams

​
Limitations

​
Current Limitations

  • URL content is not automatically refreshed
  • Processing time varies based on content complexity
  • No real-time content updates for existing sources

​
Content Processing Considerations

  • CSV hallucination risk with complex structured data
  • Processing capacity may affect upload times during peak usage
  • File size limits may apply (specific limits not currently published)
Genius is not used to provide prompts or instructions to your Voice Agent. Instead, it is used to provide detailed, unstructured information to callers. For special prompting or instructions, use the Custom Prompt setting in the Agent Builder.

​
Troubleshooting

Agent not using Genius information
  • Verify Genius is connected in Agent Builder
  • Check that content has finished processing (status: Complete)
  • Test with direct questions that match your content
  • Review Agent Binding guide
Inaccurate or inconsistent answers
  • Review source content for contradictions or outdated info
  • Simplify content structure (prefer Q&A format)
  • Remove duplicate or redundant information
  • See Best Practices for content optimization
Processing stuck or failed
  • Check file format is supported (PDF, TXT, audio, CSV, URL)
  • Verify file size isn’t excessive (contact support for limits)
  • For URLs, ensure page is publicly accessible
  • Try re-uploading the source
CSV causing hallucinations
  • Convert spreadsheet data to natural language format
  • Use simple key-value tables only
  • Consider creating text-based Q&A instead
  • See Best Practices: Structured Data

​
See also