# An AI chatbot that answers from your content, not from a model’s memory > Valenys Chatbot indexes your WordPress or WooCommerce content and answers from it using retrieval-augmented generation, showing the source of each answer. It runs on Claude, Gemini or GPT with automatic failover, has no page limit on any plan, and follows the page language through WPML or Polylang. Source: https://valenys.com/solutions/ai-chatbot/ Updated: 2026-09-08 --- _Indexed on your own WordPress or WooCommerce content, running on Claude, Gemini or GPT with automatic failover, showing the source of each answer._ A general-purpose model with no access to your actual content produces answers that are plausible and wrong. On products, prices and availability, that is worse than not answering — a confident wrong answer costs you an order and a complaint. ## What it does ### Indexes everything, with no page limit Not a sample, not the first fifty pages. The queue processes in batches with a time budget, retries with backoff, and recovers from stuck locks — so a large site on modest hosting finishes indexing instead of timing out halfway. One failed page does not block the queue. ### Answers from the indexed content, and shows where from Retrieval is hybrid: a MySQL FULLTEXT pre-filter narrows the candidate set, then cosine similarity over embeddings ranks it — which is what keeps response time usable on a site with thousands of pages. Each answer carries its source. ### Understands WooCommerce Simple and variable products alike, with SKU, price, stock and attributes; variations are indexed with the parent. Change a variation’s price, stock or SKU and the parent product reindexes — because a chatbot quoting last month’s price is worse than one that says it does not know. ### Speaks 55 interface languages A visitor-facing language selector, the interface translated automatically once per language and then cached, and right-to-left layout for Arabic, Hebrew, Persian and Urdu. On a WPML or Polylang site it follows the page language, and content is indexed separately per language. ### Captures contacts and hands over to a person Leads are posted to webhook endpoints, which covers Zapier, Make and most CRM systems directly. When the conversation needs a human, it goes to one. ### Keeps running when a provider does not Claude, Gemini and GPT can all be configured, with automatic failover. If Anthropic or OpenAI has an outage, the bot continues on another provider instead of going silent. Premium voice through ElevenLabs or Google Cloud Text-to-Speech on the higher plan. ## You can see what it answered, and what it cost Conversations, how many answers were billed and how many came free from the FAQ, response time, satisfaction, and the questions people actually ask. A chatbot nobody measures is an expense, not a tool. ## What went into the last release Versions rather than dates: a plugin ships as a number. ## Plans Plus €349 once, for setup and training on your content. Capabilities are enforced on the server, not only in the interface — a plan’s limits cannot be lifted by editing a setting. ### Start — €99 / month Full indexing with no page limit, one AI provider, the visitor widget in 55 interface languages. ### Business — €149 / month Multi-provider failover, WooCommerce indexing, lead capture, handover to a person, analytics. ### Enterprise — €299 / month White label, premium voice, custom integrations, unlimited external connections. ## What we will not claim We will not tell you it never makes things up. No supplier can honestly promise that of a language model. What we can tell you is what the design does about it: the bot answers from your indexed content, it shows the source of each answer so a visitor can check, and it reindexes when your data changes. Those three things are what reduce the risk. A guarantee would be a sentence, not a mechanism. ## Every page, every product, every variation The queue is visible: what is waiting, what failed, how many chunks came out. A chatbot that indexes the first fifty pages and says nothing about it looks exactly like one that indexed them all — until somebody asks about the fifty-first. ## Tell us the site, the size, and the languages The higher plan removes our branding entirely, and licensing is enforced per activated domain — so agencies can run it under their own name across a portfolio. ## Frequently asked questions **Is there a limit on how many pages it indexes?** No, on any plan including Start. Indexing runs in batches with retries, so site size is a question of how long the first index takes, not of whether it completes. **How does it avoid inventing answers?** It retrieves from your indexed content and answers from that, showing the source. It is not a general knowledge bot with your site attached. That reduces invention substantially; it does not eliminate it, and we would rather say so. **Does it work on a multilingual site?** Yes. It follows the page language through WPML or Polylang, indexes content separately per language, and offers a visitor-facing selector across 55 interface languages, including right-to-left scripts. **What happens to conversation data?** Retention defaults to 30 days and is configurable between one and 365 days. **Can it pass leads to our CRM?** Yes. Captured contacts are posted to webhook endpoints, which covers Zapier, Make and most CRM systems directly. Deeper CRM and ERP integrations are part of the Enterprise plan. **What if the AI provider has an outage?** Claude, Gemini and GPT can all be configured, with automatic failover on the plans that include it. If one provider is unavailable, the bot continues on another. **Does it work outside WordPress?** The licence server is platform-agnostic and already records the platform per activated domain — WordPress, OpenCart, PrestaShop, Magento, Shopify or a custom build. The WordPress plugin is what ships today; other platforms are quoted as an implementation. **Can we remove your branding?** Yes, on the Enterprise plan, with your own text and link.