Translation & Languages
enuchat translates every message in real-time so your team and visitors always communicate in their own language. 60+ languages, fully automatic.
How It Works
Translation happens automatically in both directions. Neither the visitor nor the operator needs to do anything.
The flow
- Visitor sends a message (e.g. in Spanish)
- enuchat detects the language using Claude Haiku
- The message is translated to the operator's language (e.g. English)
- The operator sees both the original and the translation in the dashboard
- The operator replies in English
- The reply is translated to Spanish and delivered to the visitor
The visitor only ever sees messages in their language. The operator only ever needs to read and write in their own language. enuchat handles everything in between.
Supported Languages
enuchat supports 60+ languages for both detection and translation, covering the vast majority of global web traffic.
Language detection
- Detection is automatic — based on the visitor's browser language combined with message content analysis
- The first message in a conversation triggers language detection
- Once detected, the visitor's language is stored for the entire conversation
- Claude Haiku provides highly accurate detection, even for short messages
Same Language Optimization
When the visitor and operator share the same language, enuchat is smart about it.
What happens
- Translation is completely skipped — no API calls, no processing
- Zero tokens are consumed for translation
- No "Translating..." indicator appears in the chat
- Messages are delivered instantly without any translation delay
For example, if both the visitor and operator speak Polish, messages flow directly without touching the translation system. This saves tokens and reduces latency.
Translation Controls
Translation is governed by three independent checks. All three must pass for translation to happen.
| Control | Level | Description |
|---|---|---|
Widget Toggle | Per widget | Master switch. When off, no messages in this widget are translated. Set in widget settings. |
Operator Toggle | Per operator | Individual preference. Operators can disable translation for themselves if they're multilingual. |
Tokens Check | Account-wide | If your account has zero tokens, translation is skipped to avoid service interruption. |
Translation Cache
enuchat caches translations to reduce token usage and improve response times.
How caching works
- Each translation is cached using a content hash (source text + source language + target language)
- Cache is cross-tenant — if another enuchat customer already translated the same text between the same languages, the cached result is reused
- Cached translations are stored for 7 days
- Cached translations cost 0 tokens
Common phrases like "Hello, how can I help you?" in popular language pairs are almost always cached, meaning zero cost for those translations.
AI Reply Translation
When AI auto-reply is enabled, the AI generates responses in the visitor's language. But operators need to understand what the AI is saying too.
How it works
- AI generates a reply in the visitor's language
- The reply is also translated to the operator's language
- Operators see both versions in the dashboard, so they can monitor AI responses
- This ensures operators always know what was communicated, even if they don't speak the visitor's language
Token Costs
Translation uses Claude Haiku for the best balance of quality, speed, and cost.
| Action | Approximate Tokens | When |
|---|---|---|
| Translation (per message) | ~1,200 tokens | Each message that needs translation |
| Language detection | ~350 tokens | First message in a conversation |
| Same language | 0 tokens | When visitor and operator share a language |
| Cached translation | 0 tokens | When an identical translation exists in cache |
Optimization details
- Language detection only happens on the first message of each conversation
- Subsequent messages skip detection if the visitor's language is already known
- Same-language conversations use zero tokens for translation — no detection, no translation calls
See the Billing guide for token pricing by plan.