Describe it.
Watch it get built.

Websites, apps and automations, built from a sentence.
A few cents to build, from $10 a month billed annually to keep online.

A beautiful place to begin

Your idea. A head start.

6 considered designs for business websites and online stores.

Explore all 6

4 templatesWebsitesMake your mark

Renew template preview: subscriptions & repeat purchases

2 templatesEcommerceTurn interest into a bag

Real pages to explore, useful sample journeys to try, and code you can make your own. See what building with Amelia costs

What you get

Everything you need, and nothing that traps you.

What the money is for — and what happens to everything you built if you ever stop paying it.

You own the code

A real Git repository, in your name, from the first build.

Take it with you

Zip it, clone it, or push it to your own GitHub. Any time, plan or no plan.

Hosting included

Your sites online on your own domain, included with every plan that builds. How many depends on the plan — the count is on Pricing.

Build from anywhere

Your project runs on a machine in the cloud, not on your laptop. Nothing to install, and it works from a phone.

Never shop for an AI again

There are dozens of AIs and the priciest costs 80x the cheapest. We send each job to the one that does it well for the least. You just build.

The price on every answer

The rate is under the composer before you send, and what it cost is printed on the answer. No credits, no packs.

Cancel any time

Your code stays yours. Hosting stops a week later, and one click brings it back.

A safety check

It tries your live site like a stranger would — leaks, broken links, slow pages.

Included usage resets each month, and credit you buy never expires. A site published on the free trial is a seven‑day preview on our subdomain; on a plan it stays up as long as the plan does. What each plan costs · everything it does, room by room.

Studio

You type. It plans, then builds.

She asks the questions, writes the plan and prices the work before anything runs. The planning turns are priced like every other turn — what the model cost us plus your plan’s markup, about $0.0012 on the model we recommend, printed on the answer. Approve the plan and the build starts on a private machine — every file it writes and every command it runs on screen as it goes, with the cost of the turn on the answer when it lands.

The whole turn, explained →

The app: a message asking for a barbershop booking site, Amelia's reply with the plan, and the line under the composer showing about $0.0033 a turn plus machine time. — ameliasagent.com/app

Library

Over three hundred models, each with its price on it.

Balanced, best result, keeps cost down — or your own computer, or pick from the full list yourself. The price is shown against every one before you spend anything, which is the thing no other builder on this market will print.

All 300+, with prices →

The model picker: keep costs down, balanced, best result, or pick from the full list — with the price shown against each. — Model & cost

Account

A spending limit, on from day one.

The fair objection to paying for what you use is that a mistake could cost a fortune. So the ceiling is on before you have spent anything, it is yours to set, and reaching it stops the next turn rather than sending you an invoice.

How the limit works →

The account page: usage included in your plan, and a spending limit that says nothing can be spent past it. — Plan & billing

Not just websites

Games, too. Play them.

Real builds, running right now — tap one and it opens in a new tab to play. Every one was described in a sentence and built by Amelia, and every score is saved to a database we host, shared across everybody rather than just this browser.

Memory, a game built with Amelia’s Agent

Memory Match every pair against the clock. Fastest time saved

2048 Slide tiles, combine numbers, hunt 2048. Best score saved

Snake The classic — grow without hitting yourself. Best score saved

Breakout, a game built with Amelia’s Agent

Breakout Clear every brick before the ball drops. Best score saved

What builds cost · how to make your own

Against the alternatives

Compare it for what you use.

Credits are a currency with no published exchange rate. Seats are a bill that arrives whether anyone built anything or not. Here the unit is a dollar — so set the usage yourself and read the answer off the chart, rather than taking ours at the one level we would have picked. Ours is the only line on any of the three charts that moves, because everybody else sells a flat subscription — and a flat bar is not “cheaper”, it is the price of a box whose contents they do not publish.

What a month of asking costs 600 answers a month

Amelia’s Chat AI $5.99 $5.99 a month · $2.50 of usage included · then $0.0012 an answer

ChatGPT Plus $20.00 flat · usage metered in five-hour windows

Claude Pro $20.00 flat · “usage limits that reset on a rolling five-hour session window”

Perplexity Pro price not published to us

If every single answer used a frontier model at $0.05 instead, the same plan would be $33.49 — the worst case, not the normal one.

What each one gives you
What you get	Amelia’s Chat AI	ChatGPT Plus	Claude Pro	Perplexity Pro
The price of the answer, printed on the answer	yes	no	no	not published
You pay for what you actually asked	yes	no	no	not published
No five-hour usage window	yes	no	no	not published
Files, photos and videos in the chat	yes	yes	yes	not published
Memory that carries between chats	yes	yes	yes	not published
Upgrade to a plan that builds and publishes	yes	no	no	no

Ours: $0.0012 an answer past the $2.50 included — the median we actually billed per turn at the Balanced posture, in the same dollars the app itself prints on every answer. Claude Pro $20 ($17 billed annually), claude.com/pricing, 6 Sep 2026. ChatGPT Plus $20, openai.com/chatgpt/pricing — that page refuses automated readers, so this is the figure already on this site, read by a person on 31 Aug 2026. perplexity.ai would not answer a fetch on 6 Sep 2026.

What a month of building costs 120 builds a month

Pro $12.00 $12 a month · $3 of usage included · then $0.009 a build turn

Pro Plus $35.00 $35 a month · $6 of usage included · a database, and your own domain

Lovable Pro $25.00 $25 for 100 credits — credits, not published per build

Replit Core $20.00 $20, “$20 towards most powerful models” — effort, not published per build

Bolt Pro $25.00 $25 for 10M tokens — tokens, not published per build

Base44 Starter $16.00 $16 for 100 message credits — credits, not published per build

Retool Business $65.00 $65 per builder — a seat, not a build

What each one gives you
What you get	Ours	Lovable	Replit	Bolt	Base44	Retool
The price of the build, before and after	yes	no	no	no	no	no
Choose the model	yes	no	not published	not published	yes	no
Your code as a git repo from the first turn	yes	no	no	no	no	no
You pay for what ran, not for credits that expired	yes	no	not published	no	not published	no
Run it yourself for $0	yes	no	no	yes	not published	no

Ours: the plan price plus $0.009 a build turn past the included usage — the heaviest real account's average, so the pessimistic figure. Replit Core $20, replit.com; Bolt Pro $25 for 10M tokens, bolt.new; Base44 Starter $16 for 100 message credits, base44.com; Retool Business $65 per builder ($50 billed annually), retool.com — all 6 Sep 2026. Lovable Pro $25 for 100 credits, lovable.dev/pricing 27 Aug 2026: that page renders its plans in the browser and would not give them to a fetch on 6 Sep, so the figure is the dated one already on this site.

What a month costs as people join 20 people

Enterprise $119.00 $119 a month flat · everybody included, no seat fee · usage at cost + 20% on top

ChatGPT Business $400.00 $20 per user, per month

Claude Team $500.00 $25 per seat, per month — $20 billed annually

Retool Business $1,300 $65 per builder, per month — and $18 for every internal user who only looks

Lovable Teams price not published to us

What each one gives you
What you get	Enterprise	ChatGPT	Claude	Retool	Lovable
Everybody at the company included, no seat fee	yes	no	no	no	not published
Adding somebody costs nothing	yes	no	no	no	not published
The price of every turn, printed on the turn	yes	no	no	no	not published
A budget per person, and a request when they hit it	yes	not published	not published	not published	not published
A price published at all, at this size	yes	yes	yes	yes	no

Ours: $99 a month billed yearly, or $119 monthly — whatever the headcount. Both are founding prices for the first hundred companies, and both are theirs permanently; what it costs after that is not set yet. Usage is billed on top at cost plus the plan’s 20%. Claude Team $25 per seat ($20 billed annually), Enterprise by conversation, claude.com/pricing; Retool Business $65 per builder ($50 billed annually) plus $18 per internal user, retool.com/pricing — both 6 Sep 2026. ChatGPT Business $20 per user, openai.com/chatgpt/pricing — that page refuses automated readers, so this is the figure already on this site, read by a person on 31 Aug 2026. lovable.dev/pricing would not render for a fetch on 6 Sep 2026.

The same six questions, asked of Lovable, Replit, Bolt, Base44 and Retool too — the full comparison. The expensive model is a choice here, not a default: the app suggests the cheapest one that does your job well, and the savings are yours.

Just want to talk to an AI that remembers you?

Amelia’s Chat AI, $5.99 a month — $4.99 billed yearly — with $2.50 of usage included and every answer priced on the answer. Files, photos and videos, memory that carries between chats, and a link to share any of them. No builds and no machine; move up to Pro when you want something put online.

Start free See the plans

Chat is charged, like everything else here · every answer shows its price

What happens

Five steps. You do the first one.

The same five whether you have written code before or not. Nothing to install, nothing to configure, no build settings to learn — everything between the sentence and the live URL is done for you.

You type what you want

Plain English. One sentence is enough to start

It asks what it needs to know

Your name, your hours, what you actually sell — each answer shows its price, and no build machine is running yet

It builds it

A private machine starts in about ten seconds. Every file, the clock and the running cost, in front of you

You look at it

It renders in the answer. Full screen at phone, tablet or desktop width

You put it online

One click, a real URL, a real certificate

Step 2 is priced per answer, with no build-machine charge. “A website for my insurance company” becomes a real brief before a single file is written. You approve the plan before building starts.

What this actually is

Every AI worth using. One sentence to drive them.

There is no such thing as one AI. There are hundreds, they are good at different things, and they cost wildly different amounts — so most tools pick one for you, never say which, and bill you in credits so you cannot tell what anything cost. Amelia picks the one that does your job well for the least, shows you the price before you spend it, and lets you overrule her. You do not have to know any of their names.

The detail, for anyone who wants it: over 300 models — American, Chinese and European, frontier and budget, each with its published price on screen before you spend anything. Seven of them we have measured ourselves and the app suggests one per job. The cheap end is not a consolation prize: on four of the five job types we measured it finished first time, and a model eighty times the price was barely better.

Or bring your own

Run the whole agent on your own hardware, with your own model keys, and we charge nothing at all. Own hardware makes the machine free; own keys make the turns free.

Your own API keys

Your account, your rate, your limits.

An open‑source model on your machine

Ollama or LM Studio — $0 per token, forever.

Your own computer as the build machine

Even while you drive it from the app. The desktop app does it in one click.

Software you already run

Point it at what you have — it reports what to cut.

Same agent, same interface, three commands to install. The hosted machines are a convenience, not a lock — run it yourself and we operate nothing that can read your code, your keys or your history. Your prompts still go to whichever model you choose, unless that model is running on your machine too — in which case nothing leaves it at all.

DeepSeek V4 Flash $0.0033

GPT‑5.6 Luna $0.0088

GLM 5.3 $0.0541

GPT‑5.6 Sol $0.0845

Claude Opus 5 $0.2113

Claude Fable 5 $0.4225

Six of the ten we measured, one average turn each, with our markup already in the number. Bars are linear, so the cheap end really is that small.

Every name here is the provider’s own — Luna and Sol are OpenAI’s GPT‑5.6 line, Opus and Fable are Anthropic’s Claude 5 line, and DeepSeek and GLM ship under those names — priced from their published rates. A 129× spread between the cheapest turn and the dearest one, on the same job. That gap is the whole reason to be able to choose, and the reason a tool that picks for you and bills you in credits can never show you this picture. Prices quoted at the trial rate, the highest we charge. See the measured ten, and the rest of the catalogue, on the models page.

Where people start

Most people arrive one of three ways.

Each of these is a real page rather than an anchor, and each one carries a measured figure — recognising yourself in a number is a faster decision than recognising yourself in a description.

You need a website

And you were about to pay a builder every month, a designer once and a developer again after that. Describe it instead, and own what comes out.
One measured build: 12 turns, $0.03 of tokens

You were quoted for an app

A booking form, a customer portal, something to replace the spreadsheet three people are fighting over. Usually not hard. Usually expensive.
53 real builds in the showcase, every price shown

You already pay for too much

Point it at software you already run. It reads the stack, names what you could own instead, says what is genuinely worth keeping, and builds the difference.
It reports before it rewrites

And if you already have something

This is not only for starting over. Point her at what you already run — a live site, a repo, a half‑finished project — and every change after that is a sentence.

The site you already run

She reads it first and reports — what is slow, what breaks on phones, what you are overpaying for. Then she makes the changes you approve, and only those.

The code you already own

Connect GitHub and import the repository. It stays private by default and she works in it commit by commit — yours before, during and after.

The context you already wrote

Paste a repo or a markdown address — an AGENTS.md, a CLAUDE.md, a project note — and its facts join her memory of your project before the first turn runs.

Writing code for a living? The same thing, with the repo, the shell and the model choice you would expect — for developers.

Where it lives

Every project is a repo you can clone.

A repo is a folder with a memory — every version of your project, every change, with a note saying what changed and when. We make one for you on the first turn, host it, and never charge for it. It is private unless you say otherwise, and you can take a full copy of it whenever you want. Nobody has to ask you to remember to export anything.

Stopping is not deleting, either: your projects and everything Amelia learned about them are still here when you come back, so starting again is not starting over. And every project has a memory of its own — your colours, your business, the decision you made three weeks ago.

Thrown away every hour

the build machine
its scratch files
whatever was half-typed in a shell

You stop paying for it the moment it stops. Nothing you care about lives here.

Kept

every file it wrote
every version, with a readable note
what Amelia learned about the project
the live site

On our git host, free, private by default — and mirrorable to your own GitHub if you would rather it sat there too.

The line between the two is the thing to check in any tool like this: what survives when the session ends, and who has to remember to save it.

Committed after every turn, with a message you can read. Plain git — clone it, fork it, push it anywhere. Never a separate bill: hosting your code is not a product here.

Not a developer and none of that means anything to you? It still matters, and somebody you hire later will know exactly what to do with it. See your code.

your laptop

$ git clone https://git.ameliasagent.com/you/my-business-site.git $ cd my-business-site $ python3 -m http.server 8000 ✓ open localhost:8000 — it is just files

Where your code actually is

On a machine we run

The sandbox is ours and the git repository is on our server in Germany. It is private to your account, and you can clone it and delete it whenever you like. We are not going to pretend otherwise.

Or never leaving yours

Run the agent yourself and the code, the keys and the history stay on your computer. We operate nothing that could read any of it. Your prompts still reach whichever model provider you pick — point it at a local model and even that stops.

Going live

A real URL, one click away.

Publishing a site is free on every plan — a real address with HTTPS, hosted on our own hardware, and you can take it offline any time. On the free trial it is a seven‑day preview; a plan keeps it online and puts it on a domain you own.

No build configuration, no dashboard, no deploy pipeline to learn. Prefer your own Vercel, Netlify or Cloudflare? It is a git repo — point them at it.

How publishing works · pointing a domain you own

Built

It exists in your repo. Nobody but you can see it.

Published

One click. What goes out is what was committed, not whatever was on the machine.

Live

A real address with a real certificate, in seconds. Send it to a client today.

Your own name

Keep the free address you are given — it is a real web address you can hand out. Point a domain you already own at it when you are ready, or buy one anywhere that sells them and we will show you what to change.

No build configuration, no dashboard, no deploy pipeline to learn. Published sites are served from our own hardware, which is why this is not a seat fee — see publishing.

Shared, not sold

The good prompts are somebody else’s.

The bill here is turns, not tokens — so the cheapest build is the one that does not start from an empty directory. Anyone can publish what worked for them, and anyone can install it into their own project before their first turn.

Browse the marketplace · start from working code · what it already knows about each job

Four kinds of thing

Published by people using it, installed in one click, free

Promptsthe wording that got it right first time

Skillsa capability the agent gains for your job

Loopswork that repeats without being asked twice

Startersa finished project copied into your repo

Browse them on the marketplace, or read what the agent already knows about each kind of job in the playbooks.

What it costs

Pay for what you use. Nothing else.

Free to try, with no card — enough to build something real and put it online, and what you make is yours whether you carry on or not. After that it is $10 a month for a site or two, or $29 for most people, with usage included. The rate is under the composer before you send; the cost is printed on the answer.

On your own computer it is free outright — every feature, your own keys, or a local model that costs nothing per token. No machine time either; it is your machine.

$0to try it, no card

$10a month, billed yearly

$0.22an hour of machine

$0on your own computer

The rest of the rate card, for anyone comparing: past your plan’s included usage a turn costs what the provider charged us plus a margin that falls as your plan goes up — 30% on the trial, Amelia’s Chat AI and Pro, 28% on Pro Plus, 22% on Pro Max, 20% on Enterprise. Every row, in a table.

Or see what it costs first

$0.13–$0.33 12–30 turns · under an hour

Cheapest model to best one — a 129× spread, and the app suggests which to use for your job. Machine time is from about $0.22 an hour on top. Quoted at the trial rate, the highest we charge. On your own machine: free.

Free to start, no card · no seat fee · no credits · a spending limit on from day one · when we cannot price something honestly, we say so. The whole price list.

A minute of machine $0.0036

An hour $0.22

A full working day $1.72

Billed by the minute you actually held it, and it stops itself when you are done. You do not need a computer of your own, and you are not paying for one while you sleep.

The included allowance is where most people will live: Pro’s $3 of usage is roughly 330 website builds and Pro Max’s $12 is roughly 1,330, counted at $0.009 a build turn — the heaviest real account’s average, so the pessimistic figure rather than the flattering one. Pro itself is $10 a month billed annually — half the outlay of one seat of a chat subscription — and this one runs 300+ models, keeps your repos and puts the result online. If all you want is the asking, Amelia’s Chat AI is $4.99 a month billed yearly.

Start with a sentence.

Free to begin, no card. Whatever you make is in a repo you can clone, whether you carry on or not.

Build something Read the setup guide

Every answer shows its price · a spending limit is on from day one · the code is yours from the first turn

iPhone · TestFlight

Also on your phone.

The same account and the same projects, in a real iPhone app. Free while it is in TestFlight.

Join the beta Talk to a human

or run the whole thing yourself

$ git clone https://github.com/RORHITD/amelias-webui && cd amelias-webui $ python3 bootstrap.py ✓ your machine, your keys, your models — free

Describe it.Watch it get built.