Smart Translate translates between eight languages and chains three translation sources into one fallback order: it first tries the DeepLX service you host (if you entered an address), then the OpenAI-compatible model you configured (streamed), and finally the MyMemory free API, which needs no configuration. Before switching to a new provider the page shows a yellow confirmation bar asking whether to send the source text to XXX and keep trying; declining stops there, and once you accept it does not ask again for the rest of the session.

Compared with Text Translation in the Query category, this tool adds the DeepLX and LLM channels plus the outbound confirmation step; for quick free translation, Text Translation is more direct.

1. 01Choose the languages— the source language can be Auto-detect (a script heuristic based on kana, Hangul, Han characters and Cyrillic, with everything else treated as English); the swap button in the middle exchanges the two languages and copies the translation back into the source box.
2. 02Choose an engine— Smart cascade (recommended) is the default; you can also pin DeepLX, OpenAI-compatible or the free MyMemory. Options you have not configured are marked “(not configured)” and cannot be selected.
3. 03Translate— click Translate; a badge above the result names the engine that actually ran, and each engine's failure reason is listed in a yellow box underneath.
4. 04Handle the cascade confirmation— when an engine fails, the confirmation bar appears; click Send and continue or Stop.

DeepLX is not configured and is skipped, so the first engine needs no confirmation

MyMemory handles about 500 characters per call

Only an OpenAI key is configured; translating 今天的会议改到明天下午三点 into English

The meeting today has been moved to 3 p.m. tomorrow. (badge: OpenAI-compatible model)

No keys configured, 1,200 characters of Chinese into Japanese

MyMemory splits the text at sentence boundaries into chunks of at most 450 characters, translates each and joins them back together

Languages

— only Chinese, English, Japanese, Korean, French, German, Russian and Spanish; auto-detect is a character-set heuristic, so mixed Chinese/Japanese text is read as Japanese because kana take priority.

MyMemory quota

— about 5,000 characters per day anonymously, after which the request fails; text longer than 500 characters is split at sentence-ending punctuation.

DeepLX

— needs a DeepLX service address you host or trust (POST JSON returning code=200); this site does not provide that service.

No cache and no history

— refreshing clears everything, and the LLM channel uses a fixed temperature of 0.2.

What it does not do

— it does not translate files or web pages, has no glossaries, and does not accept an official DeepL API key (that belongs to Text Translation).

The source text goes straight from your browser to the engine in use: your own DeepLX address, the Base URL you configured (with your key) for the OpenAI-compatible engine, or api.mymemory.translated.net for MyMemory. The cascade asks for confirmation before switching providers. This site neither handles nor stores the text or the keys. Do not translate text that contains sensitive information.