Go multilingual

Good news up front: you don't turn on multilingual. An AnyBot agent mirrors the language of whatever the customer writes, write in French, they get French; a voice note in Arabic comes back answered in Arabic, with no language dropdown to set. This recipe shows you how to prove that in the test chat, then what to actually configure so those replies stay grounded in your content, not just fluent-sounding.

What you'll build

A single agent that holds a real conversation in a second language, answering a typed question and a voice note in the customer's language, from your knowledge, plus the app interface switched into one of its 7 languages so your own team works in theirs.

Before you start

• An agent you can edit with the Web Chat channel enabled, so the test chat works, new here? Your first agent in 10 minutes.
• A question you know your knowledge covers, and the ability to ask it in a second language (a colleague or a translator is fine).
• Optional: a short document in that second language for Step 3.

This page owns the click-by-click. For the mechanics behind multilingual input, see Attachments & multimodal; for the app's languages, any user's profile settings.

Steps

1. Ask a question in another language

Open the Chat with Agent panel (a side panel on wide screens, a Chat button on smaller ones). Type a question your agent can answer, but in a second language, for example "Quels sont vos horaires d'ouverture?"

Expected result: The reply streams back in the same language you asked, grounded in your content. You changed nothing to make this happen.

2. Confirm the language follows the customer, not a setting

Ask a follow-up in a third language in the same thread. The agent switches again.

Expected result: Each reply matches the language of the message it answers. There is no per-agent language to configure. This is built-in behavior.

3. Add knowledge in your customers' languages

Fluency is free; grounding is what you configure. On the Knowledge Base tab, add a source in a language your customers use, a document or a URL to crawl. The agent can translate on the fly, but facts land more accurately when the source material already exists in that language.

Expected result: The source reaches Ready, and asking about it in that language returns a precise, grounded answer, names and figures intact, not paraphrased through English.

4. (Optional) Lock the behavior for short messages

Very short messages ("ok", an emoji) don't carry enough signal to detect a language. If your audience is mostly one non-English language, open AI Options and add an Extra instruction: "When the customer's language is unclear, reply in Spanish."

Expected result: Saved as a new version. Ambiguous turns now default to your chosen language instead of English.

5. Try a voice note, then switch the app language

Send a voice note in another language through a connected channel, AnyBot transcribes speech multilingually (Arabic included) before the agent reads it. Separately, from your profile settings, switch the app interface into another language.

Expected result: The spoken question is answered in kind, and the dashboard flips to your chosen language, with Arabic rendering full right-to-left.

How the language is decided

Customer message  ->  language detected from the text (or voice transcript)
                  ->  agent replies in that same language
   short/ambiguous?  ->  falls back to your Extra-instruction default, else English
   "let me talk to a human" (any language)  ->  deterministic backstop flags a handoff

Verify it works

• A typed question in a second language came back in that language.
• A follow-up in a third language switched again, no setting touched.
• A knowledge source in another language returned a grounded, precise answer.
• A voice note in another language was transcribed and answered.
• The app interface switched languages, and Arabic rendered right-to-left.

Troubleshooting

• Replies come back in English for short messages. That's the fallback for undetectable language. Add the Step 4 Extra instruction, or write a fuller sentence.
• It answers in the right language but the facts are vague. Your knowledge is likely English-only. Add sources in the customer's language (Step 3); the agent translates wording, not missing facts.
• A phone call is answered only in English. That's expected. The voice channel (answering the phone) is English-only today, independent of everything above. Route non-English callers to a text channel like WhatsApp.

What's next

• Configure your agent, role, style, and the Extra instructions that steer tone across languages.
• Attachments & multimodal, how voice notes, images, and documents become text the agent reads.
• Knowledge base, add and crawl sources in every language you serve.
• Connect WhatsApp, put your multilingual agent on the channel most of the world texts from.