Skip to main content

Prompt Customization

Control how your avatar responds through prompt configuration.

System Prompt

The system prompt defines your avatar's personality and behavior.

Configuration

Set via environment variable or admin panel:

# Environment
SYSTEM_PROMPT="You are a helpful assistant for..."

Or configure in Admin Panel → Prompts.

Example System Prompt

You are Alex, a friendly product specialist for Acme Corp.

Your role:
- Answer questions about our products and services
- Guide users through our documentation
- Provide accurate, helpful information

Personality:
- Professional but approachable
- Concise and clear
- Proactive in offering relevant information

Guidelines:
- Always cite sources when referencing documents
- If unsure, say so and suggest contacting support
- Stay focused on Acme Corp topics

Prompt Partials

Break prompts into reusable components:

Creating Partials

In Admin Panel → Prompt Partials:

  1. Persona - Character definition
  2. Guidelines - Response rules
  3. Knowledge - Domain-specific context
  4. Format - Output formatting rules

Example Partials

Persona Partial:

You are Alex, a knowledgeable assistant. You're friendly,
professional, and always eager to help.

Guidelines Partial:

When responding:
- Keep answers concise (2-3 sentences for simple questions)
- Cite documents when referencing specific information
- Offer to elaborate if the topic is complex

Combining Partials

Partials are combined in order to form the complete system prompt:

[Persona] + [Guidelines] + [Knowledge] + [Format]

Response Overrides

Skip LLM for known questions:

Types of Overrides

TypeDescription
ExactMatches exact input
ContainsInput contains phrase
Starts WithInput starts with phrase
Ends WithInput ends with phrase

Example Overrides

TriggerResponse
"What are your hours?""We're open Monday-Friday, 9am-5pm EST."
Contains "pricing""Visit acme.com/pricing for current rates."
Starts with "How do I contact""Email support@acme.com or call 1-800-ACME."

Hard vs Soft Overrides

  • Hard Override - Always use this response, skip LLM
  • Soft Override - Suggest to LLM, but allow adaptation

Phonetic Overrides

Control pronunciation of specific words:

Configuration

In Admin Panel → Phonetic Overrides:

WordPronunciation
FortiGate"For-tee-gate"
AWS"A W S"
API"A P I"

Usage

Phonetics are applied before text is sent to avatar synthesis.

RAG Context

Knowledge Base

Upload documents to provide context:

  1. Go to Admin Panel → Knowledge Base
  2. Upload PDF, TXT, or Markdown files
  3. Files are indexed for retrieval

Context Injection

Retrieved context is injected into prompts:

[System Prompt]

Relevant context from knowledge base:
---
{retrieved_documents}
---

User question: {user_message}

Best Practices

Keep Prompts Focused

  • One clear role/persona
  • Specific, actionable guidelines
  • Avoid contradictory instructions

Test Thoroughly

  • Try edge cases
  • Verify tone consistency
  • Check factual accuracy

Iterate Based on Feedback

  • Monitor conversation logs
  • Note common misunderstandings
  • Refine prompts over time