Skip to main content

​
Overview

Intelligence features are optional AI capabilities you can enable per agent in the Call Settings section of your configuration. Each can be enabled or disabled independently.

​
Gender Detection

Analyzes voice characteristics early in the call (first 5–10 seconds) to detect the caller’s likely gender (male, female, or unknown) and injects this context into the conversation automatically. To enable: Call Settings → User Gender Detection toggle When enabled, you can instruct your agent how to use this information in your prompt: 
When greeting users, be respectful and use appropriate language.

- If the caller is detected as male: Use "sir" as appropriate
- If the caller is detected as female: Use "ma'am" as appropriate
- If gender is unknown: Use gender-neutral language
Some jurisdictions regulate automated gender detection. Check local privacy laws before enabling.
How exactly gender context is delivered to the agent is an open question — see the open questions tracker.

​
Smart Call End

Automatically detects when a conversation is complete and ends the call gracefully, without requiring the user to explicitly hang up. To enable: Call Settings → Smart Call End toggle The LLM monitors for farewell signals (“goodbye”, “thanks, bye”, “that’s all”) and natural conversation conclusions. You can configure the behavior in your prompt: 
When you detect the conversation is complete:
1. Offer a final "Is there anything else I can help with?"
2. If the user indicates they're done, say a brief farewell and end the call.
You can also customize the end-call prompt that triggers when Smart Call End activates — a text area appears below the toggle when enabled.

​
Speaker Identification Beta

Distinguishes between different speakers on the same call and tracks who said what throughout the conversation. To enable: Call Settings → Speaker Identification toggle Useful for speakerphone calls, family or joint account calls, or any scenario where multiple people are on the line. Works best with clear audio and distinct voices.
Do not use Speaker Identification for authentication or security purposes.

​
Agentic RAG

Enables the agent to actively reason about what to search in the knowledge base, using multi-step retrieval for complex questions rather than a single lookup. To enable: Knowledge Base section → Agentic RAG toggle Requires at least one knowledge base item to be attached. Adds latency (~500–2000ms per knowledge query) in exchange for more accurate answers on multi-part or complex questions.

​
Language Dialect Switcher

Detects the caller’s Arabic dialect and adapts speech recognition to improve understanding, even if the agent is configured with a different Arabic dialect. To enable: Call Settings → Language/Dialect Switcher toggle For example, an agent configured with Egyptian Arabic can better understand a caller speaking Gulf Arabic. Detection happens once at the start of the call with minimal ongoing latency impact. This feature is designed for Arabic. English operates with a single general speech recognition model.

​
Next Steps

Write Effective Prompts

Leverage intelligence features in your prompts

Call Behavior Settings

Configure timing and interaction controls

Knowledge Base

Set up a knowledge base for Agentic RAG

Testing Agents

Test your agent before going live