I'm Johnny, the founder. Embent answers only from pages it actually crawled from your site, cites the page under each answer, and refuses when your site doesn't cover the question every refusal lands in an "Unanswered" inbox so you know which page to write next. Launching tomorrow if you're curious how the crawling, retrieval, or the "I don't know" behavior works under the hood, ask away. Blunt questions welcome.
Thank you, Necati, and great questions.
Multi-language: yes, out of the box, in the way that matters most: the agent answers in whatever language the visitor writes in, regardless of the language your site is written in, a Spanish visitor asking about your English docs gets a Spanish answer, still cited to the English page. The widget's own chrome (buttons, placeholders, the lead form) ships in 15 languages today: English, Spanish, Portuguese, French, German, Italian, Dutch, Polish, Turkish, Indonesian, Japanese, Korean, Chinese, Hindi and Vietnamese, auto-detected from the visitor's browser, or pinned per bot. Adding one more is a small data change on our side, so just tell me which one you'd need.
Infrastructure: Embent is fully hosted, nothing to install on your servers. You point it at your public website or docs, our crawler reads the pages, and each bot gets its own isolated index (no shared vector store across customers). Embedding is one script tag. There's no self-hosting option today. The conversation engine is channel-agnostic by design, so a public API and other channels: WhatsApp, Messenger, Telegram, Slack are on the roadmap; if a direct API integration is what you'd use, I'd genuinely like to hear the use case, it helps me sequence it.
Fastest way to see the language behaviour for real: paste your own URL on embent.com (no signup) and ask it something in a language your site isn't written in.