An AI voice agent that lives on your website. It listens, understands, and speaks back — using only your content. One script tag. Zero hallucination.
WebRTC voice · RAG-powered answers · Sub-500ms latency · 5KB loader
TRUSTED BY TEAMS AT
FEATURES
Real-time STT via Deepgram Nova-3, LLM reasoning through GPT-4o-mini, and Cartesia Sonic-3 TTS — all streaming over WebRTC.
Automated crawling with Playwright. Chunked embeddings via text-embedding-3-small. pgvector similarity search. The agent is grounded — it only speaks what your site says.
5KB async loader. Sandboxed iframe. Zero impact on Core Web Vitals. Works with React, Next.js, Vue, or vanilla HTML.
<script src="YOUR_WIDGET_DOMAIN/loader.js" />Live conversation streams, per-widget analytics, pinned answers, unanswered question detection, and full knowledge base management.
No mic? Behind a firewall? The widget auto-detects and degrades to text chat. Same RAG pipeline, different modality. Nobody gets blocked.
HOW IT WORKS
Drop in your URL. Our Playwright-based crawler maps your entire site structure.
Content is chunked, embedded with OpenAI, and stored in pgvector for instant retrieval.
Paste one async script tag. The widget loads in a sandboxed iframe — zero config.
Visitors speak. Your agent answers. Sub-second voice pipeline, end to end.
PRICING
Start free. Upgrade when you need more.
One script tag. Real-time voice AI. Grounded in your content. Free to start.