Multi-Language

Configuring agents to speak and understand multiple languages natively without translation layers.

View as Markdown

Iqra AI is built to be Native Polyglot.

Most platforms use a "Translation Layer" (User speaks Arabic -> Translate to English -> AI processes -> Translate back to Arabic). This causes high latency and loss of nuance.

We use a Parallel Context architecture. You define the prompt, script, and responses specifically for Arabic, and specifically for English. The agent switches its entire "Brain" and "Voice" to the target language instantly.

Configuration

1. Business Languages

First, ensure your languages are enabled at the Workspace level.

  • Go to Settings -> General.
  • Add languages (e.g., en-US, ar-SA, es-ES).
  • Set a Default Language.

2. Agent Configuration

An agent does not automatically speak all business languages. You must enable them.

  1. Go to Agent Studio -> Intelligence.
  2. You will see tabs for each Business Language.
  3. Map Providers:
    • English: STT (Deepgram Nova-2) / TTS (ElevenLabs Turbo).
    • Arabic: STT (Azure Speech) / TTS (Azure Neural or ElevenLabs Multilingual).

Optimization

This allows you to pick the best provider for each language. Azure might be better for Arabic dialects, while Deepgram is faster for English.

3. Script Content

In the Script Builder, every text field (User Query, AI Response) is multi-lingual.

  • The Global Selector: Use the dropdown in the top header to switch views.
  • Validation: Look for the checkmark next to the language name. This confirms all nodes have content for that language.

Switching Languages (Runtime)

How does the agent know which language to speak?

A. Inbound (IVR)

For Inbound Routes, we recommend an explicit selection.

  • System: "Press 1 for English. Press 2 for Arabic."
  • Result: The call starts in the selected context.

B. Outbound (CSV Override)

For Outbound Campaigns, you define the starting language in the CSV.

  • to_number: +968... -> override_agent_language_code: ar

C. Mid-Call Switching (Dynamic)

The agent can switch languages during the conversation.

  • Use Case: User starts in English but struggles, asks "Do you speak Arabic?"
  • Implementation: Use the Change Language node.
    • User Query: "Speak Arabic" / "حول عربي"
    • Action: Change Language Node (ar).
    • Result: The agent immediately swaps STT/TTS providers and loads the Arabic prompt context.

On this page