AI Agents
Language Support
Rapidflare agents answer in the language the question is asked in. There is nothing to switch on and no language to configure: the agent reads each message, works out its language, and replies in kind, whether the question arrives through the web widget, the dashboard, Slack, Discord, Discourse or email.
This ability comes from the large language models the agents run on, which read and write most widely used languages. Rapidflare adds no separate translation step. What it does add is a set of guarantees around that model behaviour, described below.
What happens when you ask in another language
| What you do | What the agent does |
|---|---|
| Ask a question in Spanish, Japanese, German or any other language the model understands | Searches your knowledge base and answers in that language, citing the documents it used |
| Switch languages part-way through a conversation | Follows the language of the latest message |
| Mention a product name, part number, error code or other technical term | Keeps it exactly as written; names and codes are never translated |
| Start typing in another language on a hub that suggests questions as you type | The suggestions appear in that language |
Answers stay grounded in your documentation whatever language they are written in. When the knowledge base does not cover the question, the agent says so in the user's language rather than guessing.
Where the knowledge base language matters
Rapidflare knowledge bases are built in English by default. When a website or documentation portal publishes the same page in several languages, ingestion keeps the English version, so each document is stored once and search never has to choose between translations of the same page.
Search itself is tuned for English content. A question asked in another language is matched against the English documentation, and the answer is written back in the language of the question. This is the configuration Rapidflare supports and tests: English documentation, questions and answers in any language.
Documentation in other languages
Ingesting documentation written in a language other than English is not yet a supported configuration. Search quality on non-English content is not guaranteed, and holding several translations of the same document in one hub produces duplicate results. If your documentation only exists in another language, talk to your Rapidflare contact before connecting it.
What stays in English
The agent's answers follow the user. The product around them does not, yet.
- The interface. Dashboard and widget labels, buttons, menus and error messages are in English.
- Hub text you write. The welcome message, input placeholder and starter prompts are shown exactly as the administrator wrote them. A hub holds one version of each.
- Progress labels. The status shown while the agent works, such as "Thinking" or "Searching the knowledge base", and the step titles in the conversation timeline are in English.
- Voice input. Speaking a question into the composer recognises English only. Read aloud uses your browser's default voice, which may not match the language of the answer.
- Citations. Links point to the original document, so a Spanish answer built from English documentation cites English pages.
Serving more than one audience
To give each audience starter prompts and a welcome message in its own language, create one hub per language and embed each where that audience lands. The agent answers in the visitor's language either way; the hub only decides what they see before they ask.
Language in analytics
The dashboard detects the language of every conversation and reports it in Analytics as a Languages breakdown, counted once per conversation. Select a language in the chart to filter the conversation list to it, which is the quickest way to see how much of your traffic arrives in each language and how those conversations went.
Related
- Rapid Support, the agent most conversations run on
- Sitemaps and Fluid Topics, which choose the English version of multilingual sources during ingestion
- Analytics for the Languages breakdown