The system instruction tells the model to identify the language of the first user message and continue in that language for the rest of the conversation. There is no separate detection step or third-party API. The model decides inline, which keeps latency low and avoids edge cases where a detector and a generator disagree.

Any language the underlying model handles. GPT-4 class models, Claude 3 family, and Gemini 1.5 all generate fluently in dozens of languages including Arabic, Hindi, Japanese, Korean, and most European languages. Quality varies by language and model, so test with the locales you actually expect on your site.

Yes. SleekAI variables resolve against translated post rows when the detected language matches a configured site locale. A product whose French translation exists in wp_posts with a matching language taxonomy gets quoted in French. If no translation exists, the bot falls back to the default locale's content but keeps replying in the user's language.

The model handles short code-switches gracefully. A sentence with a few English brand names inside a French question stays French. A full mid-conversation language switch sticks with the original detected language unless the new language dominates several consecutive turns, at which point the bot adapts.

Yes. The system instruction is editable, so you can hardcode a specific language for a bot intended for a single market. You can also use display conditions to route different bots to different URL patterns, like /fr/ getting a French-only bot and /de/ getting a German-only bot, independent of detection.

No measurable latency. The detection happens inside the same model call that generates the reply. There is no extra round-trip to a detection API. Time to first token stays the same as a single-language bot using the same model.

Modern frontier models identify language correctly even on three-word queries in most cases. Very short ambiguous messages like 'OK' or single brand names default to the previous turn's language, or to the site's primary locale if it is the first turn. You can tune this behaviour in the system instruction.

The detected language can be appended to the conversation log as a custom field, alongside the model name and token count SleekAI already records. Analytics dashboards then break down inbound chat volume by language, which is useful for staffing decisions and content localization priorities.