Crawl. Embed. Answer. Visitors speak or type — get grounded AI replies with zero hallucination. One line of JavaScript. Live in minutes.
Built for teams that ship documentation, support, and product knowledge
No models to train. No infrastructure to manage. Just paste your URL and one line of JavaScript.
Drop your website URL. We auto-detect sitemap.xml or crawl relative links — your choice.
HTML boilerplate stripped, content chunked into semantic blocks, and Gemini embeddings stored in your vector index — all async.
Paste the script tag anywhere. Visitors speak or type — your assistant answers from your own content with citations.
A complete engine — from crawler to vector store, from LLM router to live analytics — engineered for sub-50ms cache hits and zero hallucination.
Sitemap discovery, boilerplate cleaning, semantic chunking (300–500 words), and Gemini embedding generation — all running as background queue jobs.
Exact MD5 hash match → semantic vector cosine ≥0.92 → LLM fallback. Save 90%+ tokens.
Drop-in voice assistant with hands-free STT + TTS via Web Speech API.
Websites manager, sandbox tester, and live token analytics per site.
View-as-client impersonation, global sites directory, cache hit-rate matrix, queue health.
Visitors click the mic, speak naturally, and hear grounded answers read aloud — powered by HTML5 Web Speech API. No keyboard, no friction, no leaving the page.
POST /api/v1/widget/ask.
Every request is tagged with its resolution path, latency, and token consumption. Filter by site, drill into 14-day trends, and watch the live activity feed as visitors ask questions.
Paste your URL. Get your token. Embed one line of JavaScript. Your visitors will be talking to your docs by lunch.
No credit card required · 500 free queries / month