Agent structure
Each agent consists of several configuration sections:General settings
Basic agent metadata displayed in the UI.string
default:"🤖"
Visual identifier for the agent (single emoji)
object
required
Multi-language agent name
object
Multi-language description of agent purpose
Context settings
Control which business data is automatically injected into the agent’s system prompt.boolean
default:"true"
Include company branding information
boolean
default:"true"
Include branch locations and details
boolean
default:"true"
Include service catalog information
boolean
default:"true"
Include product catalog information
Personality
Define your agent’s character and behavioral guidelines. All fields support multi-language configuration.object
The agent’s persona name (e.g., “Sarah the Support Specialist”)
object
What the agent does
object
List of what the agent can do
object
Behavioral guidelines and restrictions
object
Communication style directives
Well-defined personality traits dramatically improve conversation quality. Be specific about tone, capabilities, and ethical boundaries.
Utterances
Configure opening and closing messages.enum
How the conversation begins:
None- Wait for user to speak firstStatic- Predefined greetingDynamic- AI-generated based on context
object
Multi-language static opening (if OpeningType is Static)
object
Multi-language closing message
Interruptions
Configure turn-taking and barge-in behavior. See Interruptions for detailed configuration.boolean
default:"false"
Enable strict turn-taking (no barge-in allowed)
object
required
Configuration for detecting when user has finished speaking
object
Optional: Pause agent speech when user starts talking
object
Optional: Use LLM to verify if interruption is intentional
Knowledge base
Configure RAG (Retrieval Augmented Generation) for your agent.boolean
default:"false"
Enable knowledge base integration
enum
When to retrieve knowledge:
OnEveryQuery- Search on every user messageOnDemand- Only when AI requests it via toolHybrid- Combination approach
integer
default:"5"
Number of relevant chunks to retrieve
number
Minimum similarity score (0.0 - 1.0)
object
Optional: Use LLM to refine and summarize retrieved chunks
Integrations
Connect your AI service providers. Iqra AI follows a “Bring Your Own Model” architecture.object
required
Language model configurationSupported providers:
- OpenAI (GPT-4, GPT-3.5)
- Azure OpenAI
- Anthropic (Claude)
- Google (Gemini)
- Groq
- Custom endpoints
object
required
Speech-to-Text configurationSupported providers:
- Deepgram
- Azure Speech
- Google Speech
- AssemblyAI
object
required
Text-to-Speech configurationSupported providers:
- ElevenLabs
- Azure Speech
- Google TTS
- OpenAI TTS
- PlayHT
Cache
Optimize performance with intelligent caching.Audio caching
boolean
default:"false"
Automatically cache repeated TTS outputs
integer
default:"3"
Cache after N identical generations
Embeddings caching
boolean
default:"false"
Cache vector embeddings for knowledge base
Settings
Background audio
object
S3 link to background music/ambience file
integer
Volume level (0-100)
References
Agents can be deployed to multiple channels:- InboundRoutingReferences - Phone numbers for SIP inbound calls
- TelephonyCampaignReferences - Outbound calling campaigns
- WebCampaignReferences - WebRTC/WebSocket deployments
Configuration best practices
1
Start minimal
Begin with basic personality and required integrations. Add complexity as needed.
2
Test personality changes
Small changes in tone or ethics can dramatically affect behavior. Test thoroughly.
3
Match integrations to language
Use native-language models when possible (e.g., Azure for Arabic, Deepgram for English).
4
Monitor costs
Knowledge base and LLM verification add API calls. Balance quality with budget.
5
Version control descriptions
Keep agent descriptions updated as capabilities evolve.
Next steps
Visual IDE
Build conversation scripts visually
Interruptions
Configure advanced turn-taking
Integrations
Connect AI service providers