Your AI Agent can handle calls in more than just English. Callers in Spanish, French, German, Italian, Dutch, or Japanese can be understood and answered in their own language, automatically and on the same call — no separate phone number required. This article explains how language detection works, which languages are supported end-to-end, and how to configure a multilingual agent.
How It Works
Three pieces have to speak the caller's language at the same time: the speech-to-text (STT) that hears the caller, the language model (LLM) that decides what to say, and the text-to-speech (TTS) that speaks the reply.
- STT auto-detects the spoken language. Nothing to configure per call.
- LLM is natively multilingual. It reads the transcript in whatever language the caller used and replies in that same language, as long as the system prompt allows it.
- TTS needs to switch voices to speak the target language. When the agent detects a new language, it calls an internal switch_language tool to change the voice before speaking.
The switch happens mid-call, usually within the first caller turn. The caller hears one short reply in English (or the default language), then the agent picks up in their language for the rest of the call.
Supported Languages
End-to-end (understand and speak): English, Spanish, French, German, Italian, Dutch, Japanese.
Each supported language has one or more default voices. Spanish has the widest selection with regional variants (Mexican, Castilian, Latin American, Colombian, Argentine). French, German, Italian, Dutch, and Japanese each ship with a small set of defaults that you can override per agent.
Enabling a language the agent cannot speak is fine — the fallback to a human transfer is automatic. It is still better than a confused "I did not understand" loop.
Enabling Multilingual Mode
- Open the agent edit page and go to the Multilingual tab.
- Toggle the Enabled Multilingual Support on.
- Select the default language the agent opens the call in. This is the voice callers hear before any switching happens.
- Check any additional supported languages the agent should handle.
- For each supported language, pick a preferred voice (or leave the default).
- Save. The new settings take effect on the next call.
Prompt Considerations
Your system prompt is written in one language. That is fine — the LLM can follow English instructions and still reply to the caller in French. A few tips to keep multilingual calls feeling natural:
- Keep instructions in English (or whatever you normally write in). Translating the prompt does not help; it just introduces drift.
- Spell out localized business rules if they exist. "French callers are always Quebec-based" or "Spanish callers may ask about hours in 24-hour format" prevents the agent from defaulting to assumptions.
- Test greetings in each language. Automatic translation of a scripted opener can sound stiff. Consider a short, warm opener that works natively in each language.
Troubleshooting
- Agent keeps replying in English even though the caller spoke Spanish. The language was likely not enabled in the agent's supported list. Check the Language tab.
- Agent transfers instead of speaking the caller's language. Likely Hindi, Russian, or Portuguese — STT can hear them, but TTS cannot speak them. Use a different agent or add a bilingual human fallback.
- Accent is not quite right. Swap voices. Each language has multiple voice options with different regional qualities.
Still have questions?
Submit a ticket and we will help you launch a multilingual agent.