Inbound · On-site chat

An on-site chat that answers from YOUR content — and captures emails as leads.

Retrieval-augmented, source-cited, and budget-gated. Drop a one-line script onto any site and visitors get grounded answers in seconds — while every conversation quietly turns into a qualified lead in your CRM.

Do you support self-hosting?

Yes — Growzo is open-source and self-hostable with your own API keys. There’s a one-command Docker setup and a ₹10,000/mo cap.

Sources:/docs/self-host/pricing
Can you email me the setup guide?
Send
100%
Grounded answers
Cited
Source pages
Email→lead
In-chat capture
1-line
Embed snippet
Indexed from your published content

Publish a page, and the chat learns it — automatically.

When you hit publish in the Growzo Pages editor, the page body is split into semantically sized chunks and embedded into a per-project vector index. There’s nothing to import, nothing to sync. Use OpenAI’s text-embedding-3-small for production-grade retrieval, or the free local fallback to develop and ship without paid keys.

  • Auto-indexed on publish — no manual sync step
  • Per-project vector index, isolated by projectId
  • text-embedding-3-small or free local fallback
  • Re-indexed on every page edit + re-publish
Pages indexed
128
Chunks
3,412
Embeddings
3,412
Last indexed: /pricing✓ 4s ago
Engine: text-embedding-3-small1536 dim
Grounded answers + cited sources

Every answer is constrained to what you’ve actually published.

The system prompt explicitly forbids the model from inventing facts or drawing on outside knowledge. Each turn retrieves the most relevant chunks, passes them in as context, and asks Anthropic’s Haiku to compose a tight, source-aware response. The cited page titles are returned alongside the message so visitors — and you — can audit the answer.

  • System-prompted to refuse off-content questions politely
  • Top-k chunk retrieval with cosine similarity
  • Cited page titles surfaced in the chat UI
  • Answers stay short, factual, and on-brand
How much does the Pro plan cost?

Pro is ₹4,999/month and includes the autopilot SEO loop, on-site chat with 25k indexed chunks, and HubSpot push. All plans are capped at ₹10,000/month of model spend.

Sources:/pricing/docs/limits
Email → lead at the 2nd message

Capture intent inside the conversation — not behind a popup.

The widget tracks message count per session. When message #2 lands and no email is on file, the assistant invites the visitor with a single warm sentence and renders an inline email field — no modal, no friction. On submission, Growzo upserts a Lead row, links it to the chat session for transcript context, runs the additive scoring formula, and pushes to HubSpot or your signed webhook.

  • Triggered on message 2 (configurable)
  • Lead is linked to the full chat transcript
  • Scored with the same transparent formula as forms
  • Best-effort push to HubSpot or any signed webhook
Got it — anything else I should know before signing up?
Happy to send a short setup guide over email — what address should I send it to?
Send
New lead · score 74 · pushed to HubSpot
Budget-gated, cheap by design

A chat that won’t blow up your model bill.

Answers route to Anthropic’s Haiku — the cheap, fast tier — because retrieval already does the heavy lifting. Embeddings use text-embedding-3-small (or the free local fallback). Every project has a ₹10,000/month hard cap; when you approach it, the chat degrades gracefully and posts a warning to your dashboard — it doesn’t silently keep spending.

  • Haiku for answers · text-embedding-3-small for index
  • Per-project ₹10,000/month hard cap
  • Daily spend dashboard with burn-down
  • Free local embedding fallback for dev / lean setups
Chat spend · this month₹2,840 / ₹10,000
Answers
Haiku
Embeddings
3-small
Avg / chat
₹0.42
One-line embed, drop anywhere

Ship the widget to any site in 30 seconds.

The embed is a single script tag scoped to your projectId. It mounts a floating chat button in the corner of your site, lazy-loads the conversation UI on click, and posts back to Growzo with full attribution (sourceUrl, utm, referrer). Works on every framework — Next.js, plain HTML, WordPress, Webflow, Framer — because it’s just a script.

  • Single <script> tag, no framework lock-in
  • Floating button, lazy-loaded chat UI
  • Full referrer + utm attribution captured
  • Open-source widget code, self-hostable
<script src="https://yourapp.com/api/chat/<projectId>/widget.js"></script>
yoursite.com
›_
How it works in 60 seconds

From publish to captured lead — in four steps.

01
Publish

Write or auto-draft a page in the Growzo CMS and hit publish.

02
Auto-index

The page is chunked and embedded into your project’s vector index — instantly.

03
Visitor asks

The chat retrieves matching chunks and answers with cited sources.

04
Lead captured

On message 2, an inline email field appears — the lead lands in your CRM.

Engines used under the hood
Anthropic Haiku · answersOpenAI text-embedding-3-small · indexLocal fallback · BYOK-light dev
Why teams pick Growzo for chat

A chat widget that’s actually a growth channel.

Zero hallucinations by design

Answers are constrained to retrieved content. If the index doesn’t cover a question, the assistant says so — never makes it up.

Cited sources build trust

Every answer shows the source pages it came from. Visitors can verify, and you get organic clicks back into your own content.

In-chat lead capture

The 2nd-message email prompt feels like a conversation, not a popup. Conversion rates beat sidebar forms by a wide margin.

Chat questions

What does the chat answer from?

Only your published content. When you publish a page in the Growzo CMS, the body text is chunked and embedded into a per-project vector index. Every visitor question is answered by retrieving the top-matching chunks from that index — the model is system-prompted to refuse to answer from outside knowledge.

Can it hallucinate?

The retrieval-augmented design makes hallucination very unlikely. The system prompt instructs the model to answer strictly from the retrieved context and to say it does not know when nothing relevant comes back. Every answer also surfaces the source page titles it used, so you (and the visitor) can verify.

How is the email captured?

On the second message in a new chat session, if no email is on file, the assistant warmly invites the visitor to leave one — framed in the conversation, not as a popup. On submission we upsert a lead, link it to the chat session for full transcript context, score it, and best-effort push to HubSpot or any signed webhook you configure.

Does it work without an OpenAI key?

Yes. The embeddings step has a free local fallback that produces deterministic vectors for development and BYOK-light setups. For production we recommend text-embedding-3-small (cheap and high-quality) — but you can run the whole loop without an OpenAI key.

Turn your content into a 24/7 sales chat.

Publish once, get cited everywhere, and capture every email — all under a ₹10,000/month hard cap.

Open-source · Self-host · BYOK · ₹10k/mo hard cap