DocsMultilingual

Multilingual

Alpheris automatically detects the language of every customer message and replies in the same language — no configuration, no separate bots per market. The page or path language (e.g. /en) additionally feeds in as a hint.

How it works

You do not need to enable multilingual support. As soon as a visitor writes, Alpheris determines the dominant language of the message and phrases the reply in exactly that language. A German customer gets a German reply, a French customer a French one — within the same chat, without the visitor having to change anything.

Language detection works on two levels:

The message text — the primary and decisive source. What the visitor actually writes determines the reply language.
Page/path language — the language of the embedded page (e.g. an /en path or the page's lang attribute) serves as a hint when the message text alone is not clear enough.

Nothing to set up. Language detection and replies in the detected language work out of the box for every shop and every plan. You don't need a second agent or your own language configuration.

Detecting the message language

The reply language is determined in a fixed order — from the most reliable to the fuzziest source:

Explicit language code

If a specific language code is passed with the request (e.g. via the widget field language), it takes precedence.

Script and word detection (deterministic)

Non-Latin scripts (e.g. Chinese, Japanese, Korean, Arabic, Hebrew, Cyrillic, Greek, Hindi, Thai) are recognized directly from the script; Latin languages from typical signal words. This is fast and costs no model call.

AI language detection

If the text remains ambiguous, the language model determines the dominant language of the message. For mixed texts, the predominant language is chosen.

Fallback language

If nothing definitive can be determined, the preset fallback language applies (German by default).

The AI agent's reply is then generated in the language determined this way. This applies not only to the generated replies, but also to system messages such as limit or fallback notices — these too appear in the detected language of the conversation.

Detection works per message. If a visitor switches languages mid-conversation, Alpheris uses the new language for the following reply.

Page/path language as a hint

Many shops serve the same page under language-specific paths — for example /en/produkte for the English version. Alpheris reads a leading language-segment code from the page URL (e.g. /en, /fr or /en-us) and uses it as a hint about the expected language.

Important: the path hint is not stronger than the actual message text. If a visitor on an /en page writes a German message, Alpheris replies in German. The hint comes into play mainly when the text alone is too short or too ambiguous to determine the language reliably.

What is evaluated as a language hint

In addition to the path, we evaluate — where transmitted by the widget — further signals, in this order:

Signal	Example	Origin
Explicit code	`language` / `locale`	Set by the widget or your embedding
Page language	`page_language`	The page's `<html lang="…">`
Browser language	`accept_language`	Language of the visitor's browser
Path locale	`/en`, `/fr`, `/en-us`	Extracted from the page URL

By default you don't need to set any of these manually — the widget passes the page URL automatically, and the path locale is derived from it.

Integration

There is no separate installation for multilingual support. The normal widget snippet is enough; language detection is part of the runtime.

HTML

<!-- Alpheris Chat Widget -->
<script src="https://alpheris.com/widget.js" data-alpheris-key="INSTALL_ID" crossorigin="anonymous" async></script>
<!-- End Alpheris Chat Widget -->

Replace INSTALL_ID with your install key from the dashboard. More on this under Widget Code.

If you embed the same snippet on a language-specific page (e.g. under /en), the path is automatically recognized as a language hint. There's no need for a separate configuration per language variant.

Knowledge base and language

The answers remain grounded in the content of your domain (see Data Sources & Knowledge Base). If Alpheris replies in a different language than your source texts, the information found is transferred in meaning into the visitor's language for the reply. To keep the facts accurate, it's worth having your most important language versions crawled — so add the respective language paths of your site to the knowledge base.

The less source text there is in a language, the more the reply relies on translation instead of original content. For markets with their own website translation, we recommend covering the respective language pages in the knowledge base.

Inbox & Email

Automatic language detection applies not only to the web chat, but also to the Inbox. Customer inquiries via email or messaging are answered — or generated as a reply suggestion — in the language of the incoming message. The tone follows the respective channel profile (email formal, web chat balanced, WhatsApp/Instagram friendly and brief).

Limits & notes

Very short inputs (single words, pure numbers) provide little signal. Here the page/path hint helps; otherwise the fallback language applies.
Mixed-language messages are reduced to the dominant language — the reply comes in one language, not mixed.
The fallback language is German by default and only applies when no detection is possible.
Translated replies are only as accurate as the underlying source texts — see the note on the knowledge base above.

Questions about configuration or a specific market? Write to us via console.alpheris.com or take a look at the FAQ.