Languages & cost tracking
This page covers two things that happen automatically: replying in the visitor's language, and tracking what your bot costs.
Multilingual replies
Promptly detects the visitor's language and replies in it — no setup required. The bot reads the first part of each visitor's first message to decide the language, then locks it in for the rest of that conversation so it never switches mid-chat. It also matches the right script (for example Latin vs. Cyrillic).
Detection handles common European languages including English, German, French, Italian, Spanish, Turkish, and Serbian (both Latin and Cyrillic). Very short or emoji-only messages ("ok", "👍") fall back to your account's default language instead of guessing.
The underlying search uses a multilingual embedding model (bge-m3), so a visitor can ask in one language about content you wrote in another and still get relevant results. See How retrieval works.
🖼️ [Image] — The same bot answering the same question in three different languages.
You can set a default language for your account, used for canned messages and as the fallback when detection is uncertain. You don't need to write language instructions into your system prompt.
Cost tracking
Every AI message consumes tokens, which cost money. Promptly meters this against your plan's monthly allowance.
- Monthly spend — usage accumulates against your plan's monthly message and token limits. When you approach or hit the limit, the bot is gated unless overage billing is enabled.
- Overage — if your administrator enables overage billing, messages beyond your included allowance are billed at a per-message rate instead of being blocked. See Billing & overage.
- Allowances by plan — higher plans include far more monthly messages and tokens. See Plans & limits.
🖼️ [Image] — A usage card showing messages used against the monthly allowance.
Keeping costs down
- Use and a modest Max Tokens for most bots — see Model & behavior settings.text
gpt-4o-mini - Keep your system prompt short; it's billed on every message — see Writing a system prompt.
- Lower Max History if visitors rarely reference earlier turns.
Detailed per-conversation and usage figures live in your analytics. Note that raw token and dollar cost-of-goods figures are kept internal; your dashboard surfaces usage against your plan rather than provider costs.