Language Wrappers

Language wrappers let your assistant respond natively in African languages. They work by prepending specialized instructions to your system prompt. Language Wrapper Selection

Supported Languages

LanguageCodeDescription
YorubayoWest African tonal language, 45M+ speakers
HausahaMost spoken language in West Africa, 80M+ speakers
IgboigSoutheastern Nigeria, 45M+ speakers
Nigerian PidginpcmLingua franca of Nigeria, 75M+ speakers
FulaniffSpread across West/Central Africa, 40M+ speakers
Auto-DetectautoAutomatically adapts to the caller’s language

How to Enable

  1. Open your assistant in the editor
  2. Find the Language dropdown (in the voice/transcriber section)
  3. Select your target language
  4. Write your system prompt as normal
  5. Click Save
Selecting a Language Wrapper

How It Works

When you select a language, a specialized prompt prefix is automatically prepended to your system prompt on save.
1

You write your prompt

You are a customer support agent for Acme Corp.
Help customers with their orders.
2

Language wrapper is prepended

You are a Yoruba realtime voice assistant.

Rules:
- Prefer Yoruba in your replies.
- If the transcription is noisy, mixed, or incomplete,
  infer the user's intent from context before answering.
- If the user switches language, follow them.
- Keep replies short, natural, and speech-friendly.

---

You are a customer support agent for Acme Corp.
Help customers with their orders.
3

Assistant responds in the target language

Caller: “Hello, e se, what is my balance?” Assistant: “Se o n bere nipa balance re ni?”

What Each Wrapper Does

Every language wrapper instructs the AI to:
  • Prefer the target language in responses
  • Handle noisy transcription — infer intent from context instead of refusing
  • Follow language switches — if the caller switches to English, follow them
  • Keep replies speech-friendly — short, natural, not robotic
  • Ask clarifying questions — when meaning is unclear, ask instead of guessing

Auto-Detect Mode

The Auto-Detect wrapper is multilingual. It:
  • Listens to the caller’s utterance and infers the language
  • Supports all listed languages + English
  • Doesn’t lock onto a language too aggressively
  • Follows the caller if they switch mid-conversation
Use Auto-Detect when your callers might speak different languages. Use a specific language when you know your audience.

Disabling

Select None from the language dropdown to remove the wrapper entirely. Your system prompt will be used as-is.

Best Practices

If you select Yoruba as your language wrapper, ensure your transcriber is also set to a provider that supports Yoruba (e.g., Spitch).
The wrapper already handles language behavior. Your prompt should focus on the task, not language instructions.
Use the Test button to speak to your assistant in the target language and verify it responds correctly.