A five-chapter walkthrough — non-technical, no jargon

How AI builds
your website
and why it costs so little.

"AI-built" sounds like magic, and magic is usually a scam. So here's the opposite: a plain-English walkthrough of exactly how our RAG system turns a description of your business into a custom static website — and why the smarter it gets, the less you pay.

Talk to us
$0
to see your fully custom first build
Seconds
from description to first build
100%
previewed by you before launch
Scroll
📋

Templates all look alike

DIY builders give every salon and every plumber the same five layouts.

💸

Agencies cost thousands

$3,000–$10,000 for a brochure site is real money for a small business.

DIY eats your weekends

You run a business. Dragging boxes around a page builder isn't the job.

📉

SEO is an afterthought

Most small-business sites never rank because the basics were never wired in.

Every site we ship solves all four — by design, not by luck.

Chapter 01 · Strategy

The smarter our retrieval,
the cheaper your website.

Generating a whole website with a frontier AI model on every request would be expensive — and you'd pay for it. We invert the equation: invest in retrieval (RAG). The system keeps a curated library of proven design patterns, copy strategies, and industry knowledge. The AI doesn't invent your site from nothing — it retrieves what already works for businesses like yours, then customizes it for you alone.

Generation cost collapses
Grounded in retrieved patterns, the AI does less guessing and less expensive work — which is why you start free and publish from $7/month, not $5,000.
Quality climbs
The knowledge lives in a curated library we improve every week — not frozen inside a model.
Custom, not templated
Retrieval finds what works for your industry; generation writes it for your business only.
"The richer the retrieval layer, the cheaper — and better — every website becomes."
Cost curves · as RAG matures
Total cost ↓ even as retrieval investment ↑
Why $7/mo works
$$$$$ 100 80 60 40 20 $ 0 Relative monthly cost None Stage 0 Basic keyword + docs Vector RAG embeddings Hybrid + Rerank curated KB Agentic RAG tools + memory Mature Stage 5 RAG investment / maturity → Frontier LLM (top-tier model) Fine-tuning / re-training RAG infrastructure Total cost ↓ Crossover · cheap models suffice LLM API cost Fine-tune / training RAG investment Total spend
LLM API cost Fine-tune / training RAG investment Total spend
Chapter 02 · Architecture

One system, layered:
governed data in, auditable sites out.

Your business description doesn't get pasted into a chatbot. It enters a layered, governed pipeline: an ingestion layer that structures what you told us, a retrieval layer that finds matching design and copy patterns, a generation layer that writes your pages, and an orchestration layer that logs every step. Same architecture serious AI products use — applied to your website.

AI Agent Conversational, multi-tool reasoning over enterprise data PULL · on-demand AI Automation Scheduled, no-prompt insights delivered to consumers PUSH · scheduled Business user asks a question in natural language </> AI application Secure chat interface with sign-in and app-level access controls. Agent orchestrator Routes tools, manages context, and composes the final answer. Vector search Retrieves relevant business context from approved knowledge. Structured data tool Turns questions into governed queries and table results. BI and app tools Connectors call approved APIs, semantic models, and reports. Scheduled trigger Runs daily, weekly, or when new business data lands. Automation agent Queries models, summarizes movement, and flags anomalies. Daily briefing Narrative summary, key metrics, and action prompts. no user prompt required
01
Approved data only
The AI works from your description and our curated pattern library — not from random internet guesses.
02
Tools, not magic
An orchestrator routes each step — design selection, copywriting, code generation — to the right tool.
03
Every step logged
Each site carries a build trace: what was retrieved, what was generated, who reviewed it.
Chapter 03 · The pipeline

What actually happens to
your description.

Not magic — three deliberate steps. Retrieval finds the design patterns and copy strategies that fit your industry, generation writes your pages from them, and you preview everything before it publishes. Every step is logged. Every page is traceable.

Inside one conversation — how the agent answers User asks → Vector DB retrieves context → Query engine runs the SQL → table returned to user your-ai-app.example.com/chat Ephemeral New Chat show me 10 orders placed yesterday Sure! Let me start by getting today's date and searching for the relevant business context at the same time. get_current_time Completed PARAMETERS {} RESULT 2026-05-26T20:43:45.806682 search_business_context Completed PARAMETERS { "question": "how to filter orders by date" } Ask a question... Today is 2026-05-26, so yesterday is 2026-05-25. Now let me query the data warehouse for 10 orders from yesterday using the retrieved context. query_data_warehouse Completed PARAMETERS { "question": "Show me 10 orders placed yesterday (2026-05-25). Filter usin…" } "01f159439a991a89a218c53d8c7fa29f"} Here are 10 orders placed yesterday (2026-05-25): # Order ID Customer Date Region Channel Status 1 ORD-12907 Acme Corp 2026-05-25 North Web ✅ Shipped 2 ORD-12849 Globex Ltd 2026-05-25 South Mobile ✅ Shipped 3 ORD-12846 Initech 2026-05-25 East Web ✅ Shipped 4 ORD-12943 Umbrella 2026-05-25 West Partner ✅ Shipped 5 ORD-06039 Stark Inc 2026-05-25 Central Web ✓ Verified …5 more rows below 1 VECTOR DB 2 QUERY 3 ANSWER One question, two retrievals: Vector DB finds the right context, the query engine turns the question into SQL, and the user gets a clean table back.
1 · Retrieve
The system finds the proven patterns that match your industry — layouts, sections, copy angles, color logic.
2 · Generate
AI writes your headlines, page copy, and production HTML/CSS — for your business, from your own words.
3 · Preview & publish
You preview every page on any device, refine by talking, then publish the finished static site to yourname.vierapp.com.
"

We don't ship black boxes. Every page in your site has a build trace, every retrieval has a score, and every site we launch gets to make the next one smarter.

Chapter 04 · Continuous learning

Quality compounds.
Every site makes the next one better.

A closed loop, not a one-shot generator. Every build is scored against quality checks — speed, SEO, readability, mobile layout. What works gets folded back into the pattern library; what doesn't gets fixed at the source. Your site benefits from every site that came before it.

AI Chatbot RAG Quality Training Loop Five tables that power evaluation, knowledge, serving, and tracing WHAT THIS LOOP DOES Every conversation feeds back into training. The system learns from real questions, and quality scores rise over time — measurable, auditable, never a black box. 01 Case Eval Training set of gold questions Gold questions Expected facts SQL + retrieved IDs GOVERNED DATA TABLE agent_eval_cases 02 Eval Engine Quality scoring & experiment tracking Scores traces Checks retrieval Compares answers EVAL EXPERIMENT eval_run_2686038 03 Knowledge Business context the agent reads Domain rules Query patterns Business definitions VECTOR INDEX vector_db_knowledge 04 User Question Agent orchestrates the answer Agent retrieves context Refines the question Calls the query engine AI APPLICATION your-ai-app 05 Log Table End-to-end trace of every turn Trace ID + activity IDs + scores + SQL Errors + payloads GOVERNED DATA TABLE agent_debug_log add recurring questions back as eval cases PENDING FOR GO LIVE Before we ship to all users Four open workstreams to close out a Productionalize Hardening · monitoring · rollout b Eval Training Iterate eval cases · raise quality scores c Updated Look & Feel UI polish · branding d User Acceptance UAT sign-off · feedback Quality & Continuous Learning
📊
Measurable
Quality scores tracked per build — page speed, SEO checks, mobile layout, copy quality.
🔁
Self-improving
Patterns that perform — pages that load fast and rank — get promoted in the library for the next build.
🔍
Auditable
Every build trace is stored. If something's off, we can see exactly which step produced it — and fix it for everyone.
Chapter 05 · Economics

Know exactly why
it costs so little.

An agency's $5,000 invoice pays for human hours: design, copy, code, project management. In our pipeline those are the cheap steps — and retrieval keeps making them cheaper. We track every cost driver, pull the five levers that flatten the curve, and pass the difference to you. That's the whole trick: free to start, from $7 a month to publish, no catch.

Cost — where every AI dollar goes Five cost drivers in a modern AI stack — and the levers you control 01 Query Engine Natural-language to SQL — billed by query compute COST DRIVER Query compute units + Query complexity + Idle / warm-up overhead KEY LEVERS Per-question billing · result caching · serverless vs dedicated compute 02 Foundation Models Right-sized model per task · pay only for what you use USAGE PROFILE PRODUCTION Balanced model quality & cost optimized DEVELOPMENT Multi-model bench top-tier · open-source · enterprise KEY LEVERS Pay-per-token vs provisioned throughput · cost-aware routing 03 Vector Search Embedding index — endpoint + storage + query volume ENDPOINT TYPE Standard always-on · fastest queries Storage-optimized cheaper · larger indices KEY LEVERS Right endpoint sizing · re-embedding cost · refresh cadence 04 Compute App · jobs · clusters · query engines — compute units + cloud VM WHERE THE METER RUNS App compute always-on agent Job clusters eval / refresh runs Query engine on-demand SQL KEY LEVERS Serverless vs dedicated · auto-scale · idle timeout · spot for jobs 05 T Tokens End-to-end usage across every conversation PER CONVERSATION Input retrieval + prompt Output answer + reasoning KEY LEVERS Per-user token budgets · conversation-level visibility · audit THE GOAL best-practice playbooks, billing metadata, and dashboards that attribute every $ back to its driver. Cost Visibility & Governance
Why it's different

Built like real software.

Not a template marketplace, not a drag-and-drop toy. Six things every website we ship gets right — out of the box.

🎨
Custom by default
Retrieval-grounded generation. Your site is written for your business — never a fill-in-the-blanks template.
📈
Cost goes down with use
Smarter retrieval → cheaper builds. The spend curve bends in your favor, and the price reflects it.
👁️
You approve every page
Nothing goes live unchecked — you preview every version and publish only when it's right.
Static & fast
Pure static pages on a global edge network. Under a second to load, nothing to hack, nothing to crash.
🔍
SEO wired in
Semantic HTML, structured data, sitemaps, fast load times — the ranking basics, done correctly from day one.
🔓
No lock-in
Your content is yours. Standard HTML and CSS — portable anywhere, anytime.

Want a website built this way?

The self-serve app is launching soon. Until then, tell us about your business by email and we'll start the conversation — free to start, from $7/month to publish.

Start your website Coming soonGet notified at launch
No funnel, no upsell calls. Just a website that works.