🧠 Hypertranslate AI — a second engine

A new AI translation telephone is live at /ai:

• Every hop is a prompt to a large language model hosted by Pollinations.AI instead of a Google Translate lookup — the text is re-expressed phrase by phrase, so idioms and tone hold on a little longer before the drift takes over.
• Four models are available above the form: GPT-5.6 Luna (default), GPT-5.4 Nano, Meta Llama 3.3 70B, and DeepSeek V4 Flash.
• The whole run travels over one WebSocket connection, and each round streams back token by token — you watch the route land in real time instead of waiting for a final answer.
• Switch & resume: if a hop fails or the socket drops mid-round, press Continue — finished rounds are kept and only the remaining hops re-run. You can switch models mid-run; your choice is remembered in your browser.
• Auto-detect source language, live translation route with expand/collapse, per-hop copy, and local history are all on board, alongside share links with one-tap sharing to X, WhatsApp, and Telegram.
• Full page copy, FAQ, and JSON-LD (WebApplication + FAQPage) ship in all seven locales: English, 简体中文, 繁體中文, العربية, Bahasa Indonesia, Bahasa Melayu, and Português (BR).

🌐 Language settings

A new settings page at /languages puts you in charge of the language pool:

• Toggle any language on or off — disabled languages disappear from the target language picker and no longer appear as random stops in the relay.
• Search, “only show enabled” filter, and enable/disable all make managing the list quick; at least one language must stay enabled.
• Source language is unaffected — you can still translate from any language, including disabled ones.
• Saved in your browser only (localStorage, no account), and the setting applies to both engines: the classic Google version on the homepage and Hypertranslate AI.
• Available in all seven locales.