Configure AI agents to handle multiple languages
Set up your AI voice agent to detect caller language and respond in their preferred language automatically during calls.
Centerfy voice agents can communicate in multiple languages within a single agent. Instead of creating separate agents for each language, you add language instructions to your prompt and the AI detects the caller’s language in real time, responding naturally in the same language throughout the conversation.
How language detection works
When a caller speaks, the AI agent identifies the language within the first few seconds of the conversation. Once detected, the agent switches to that language for all subsequent responses — including greetings, qualification questions, and booking confirmations. The agent maintains the same persona and follows the same conversation flow regardless of language.
Configure multi-lingual support
- Open your agent in the Agent Builder
Navigate to AI Agents and click the voice agent you want to make multi-lingual.
- Add language instructions to your prompt
In the Global Prompt tab, add a language section to your system prompt:
LANGUAGE INSTRUCTIONS: You are fluent in English, Spanish, and French. Detect the language the caller is speaking and respond in that same language for the entire conversation. If the caller switches languages mid-conversation, switch with them. If you are unsure which language the caller is speaking, default to English and ask: "Would you prefer to continue in English, Spanish, or French?" - Translate key phrases
For critical parts of your script (greetings, compliance disclosures, closing statements), include translations directly in the prompt so the agent uses your preferred phrasing:
GREETING: - English: "Hi, thanks for calling Acme Clinic! How can I help you today?" - Spanish: "Hola, gracias por llamar a Acme Clinic. En que puedo ayudarle hoy?" - French: "Bonjour, merci d'avoir appele Acme Clinic. Comment puis-je vous aider?" - Set the primary language in agent settings
In the Call Settings tab, set the primary language to your most common caller language. This helps with initial speech recognition accuracy before the agent detects the caller’s language.
- Test each language in Voice Lab
Open Voice Lab and test a full conversation in each supported language. Verify the agent maintains the correct language throughout, uses natural phrasing, and follows the conversation flow in every language.
Prompt example
IDENTITY:
You are Maria, a bilingual receptionist for Acme Health Clinic.
You speak English and Spanish fluently.
LANGUAGE RULES:
- Detect the caller's language from their first sentence
- Respond in the same language for the entire call
- If the caller speaks a language you don't support, politely say in English:
"I apologize, I currently speak English and Spanish. Would either of those work for you?"
- If the caller code-switches (mixes languages), respond in whichever language
they use most
CONVERSATION FLOW (applies in both languages):
1. Greet the caller
2. Ask what service they need
3. Collect name, email, and phone
4. Book an appointment
5. Confirm and close
How many languages can a single agent support?
There is no hard limit. However, agents perform best with 2-4 languages. If you need support for many languages, test each one thoroughly and consider creating region-specific agents for the best quality.
Does the agent's selected voice affect language quality?
Yes. Some voices handle certain languages better than others. Test your chosen voice in Voice Lab across all target languages and switch voices if the pronunciation or natural flow is lacking.
Can the agent handle language switching mid-call?
Yes, if instructed in the prompt. Add an instruction like “If the caller switches languages, switch with them immediately.” The agent will adapt within the same conversation.
What if a caller speaks an unsupported language?
Add a fallback instruction to your prompt telling the agent how to handle this gracefully — for example, apologize and offer to transfer to a human who speaks the caller’s language.