A chat, an inbox, a knowledge base, an MCP server —
one grounded pipeline.
Every answer the widget streams, every draft your inbox suggests, every article the gap-finder writes, every tool an external LLM calls — all routed through the same retrieval, the same citations, the same audit trail.
A messenger, not a bubble.
Home space with grounded suggestion chips pulled from your own content. Chat space with phased typing, inline [n] citations, and a resume card for returning visitors.
- Home → Conversation, not a cold thread
- Suggestion chips auto-generated from ingested content
- “Pick up where you left off” card with a Gemini-written summary
- Unread badge on the launcher when the assistant posts
- InlineAsk block for any page — greeting + chips that open the chat
- Visible human handoff with named agent + reply ETA
Every claim linked. Every source visible.
URL, sitemap, or PDF/DOCX/MD upload → HTML-aware chunking with section breadcrumbs → Gemini embeddings → Qdrant retrieval → BGE reranker → Gemini 2.5 Flash streaming generation. When the KB doesn't cover a question, we admit it and tag miss_reason — which feeds the gap-finder below.
It learns what you haven't answered yet —
and drafts the missing article.
Ingest URLs, whole sitemaps, or upload PDF/DOCX/MD. Every hour the gap-finder clusters visitor questions your KB couldn't answer and drafts a markdown article for each topic — ready for one-click publish.
Ingest anything
URL, sitemap.xml bulk-crawl, PDF/DOCX/MD up to 20 MB. Originals stored privately in our object store.
Section-aware chunks
HTML h1/h2/h3 breadcrumbs survive into every chunk so retrieval ranks by topic, not paragraph soup.
Gap-finder
Hourly clusterer turns unanswered questions into draft articles. Review, edit, publish — then the next visitor gets an answer.
Your team replies.
Chatified drafts.
Open a conversation. Click Suggest draft. The same grounded RAG pipeline that answers visitors produces a reply-ready draft with citation chips. Edit or send as-is — the reply lands as an assistant message tagged authored_by: "agent" and fires your webhook the same way a RAG-generated answer does.
Full transcript
Every message, every citation, every miss-reason tag — visible to your team.
Grounded draft
One click. Same retrieval. Attribution shown next to each reply.
Clean handoff
Escalation fires a conversation.escalated webhook so your helpdesk stays single-source-of-truth.
Tools, not just docs.
Chatified speaks Model Context Protocol on both sides:
- As a client: point the widget at your merchant MCP server (HTTP, JSON-RPC 2.0) — orders, CRM, whatever. We introspect your tools, you enable per-tool, we audit every call.
- As a server: npx @chatified/mcp drops Chatified into Claude Desktop, Cursor, or Zed. Your LLM can search your KB, ingest URLs, read conversations.
"mcpServers": {
"chatified": {
"command": "npx",
"args": ["-y", "@chatified/mcp"],
"env": { "CHATIFIED_API_TOKEN": "ct_live_…" }
}
}tools/list, you pick which to enable, we audit every tools/call.See the gaps. Close them.
Six panels under /app/analytics: funnel, miss-reasons, traffic sources, top-pages-driving-chats, new-vs-returning, daily volume. First-touch UTMs captured by the widget on page one, joined straight to conversation counts.
Honest about what's shipped today.
Things we do today — not certifications we're angling for. The list below is what's running in production right now.
Ready to
ship it?
Install a script tag, point it at your docs. You'll have grounded answers before your coffee goes cold.