Tune every letter
Font, size, weight, leading and spacing, all by hand.

Upload your docs. The AI answers questions grounded in your content, with citations. Skip Intercom Fin's $0.99 per resolution.
01 / What you get
Click Generate from the box above. Real Next.js code, a real database, real integrations, not a sandbox. Each piece below ships in the first prompt.
Streaming chat UI with markdown rendering and code-block syntax highlighting
Knowledge base ingestion: upload PDFs, paste text, crawl URLs, all chunked and embedded
RAG retrieval: every answer cites the source documents it pulled from
OpenAI or Anthropic backend, swap models without changing the UI
Conversation history per user, searchable, exportable
Token usage tracking per conversation so you can audit cost per user
02 / Variants
Pixel-perfect control
Font, size, weight, leading and spacing, all by hand.

Collections and CMS that feed your live pages.


Save any section and drop it in anywhere.

Pull in free imagery without leaving the canvas.


03 / Who it's for
Integrations
Payments, database, AI, email and analytics, wired up the moment you connect.







04 / Why generate it
Intercom Fin charges $0.99 per resolution. At 1,000 questions a month that's $12,000 a year and you don't own the conversations or the model choice.
You pick the model (Claude, GPT, open-weight via Ollama later), the embeddings (OpenAI, Cohere, local), and the vector store (pgvector, Pinecone). When prices drop or a better model ships, you swap one line.
FAQ
Everything you need to know before you generate it. Anything else, our team is one email away.
Upload PDFs or paste URLs into an admin page (Webtwizz scaffolds the form). The AI scaffolds a route that fetches/parses the content, chunks it into ~500-token segments with overlap, and stores embeddings in pgvector (Postgres extension via Supabase). Re-ingestion is incremental, only changed docs re-embed.
RAG is right for almost everyone. Fine-tuning is expensive, slow, and brittle when your docs update. RAG retrieves fresh context per query, which is why the answer can include "this changed last week" without retraining. Keep prompt engineering tight: instruct the model to refuse if retrieval is empty.
Yes. The AI scaffolds citations inline, each answer ends with a "Sources:" block listing the doc titles and URLs the model pulled from. Clicking a citation deep-links to the source. Trust signal users actually look at.
Roughly $0.01-$0.05 per conversation on Claude Haiku or GPT-4o-mini for typical Q&A length. Embeddings are ~$0.0001 per chunk, one-time at ingestion. So a 100-page doc set costs cents to embed; 1,000 user conversations a month costs $10-$50 in API spend. Compare to $990 per 1,000 resolutions on Intercom Fin.
Yes. Webtwizz scaffolds an embed snippet (a script tag + iframe) that renders the chat in any website. Auth is optional, anonymous users get a session ID; logged-in users get history attached to their account.
Keep building
Start building
Describe what you want. WebTwizz builds the real thing, then you click to change anything. No code needed.
Get started for free, no credit card needed.