A machine per session
Real shell, real filesystem, real processes — see hosted machines, or pair your own computer and pay us nothing for it.
AI coding agent
An agent with its own machine.
Not autocomplete and not a chat window — an agent that plans, writes, runs and fixes real code on a real computer, and commits every turn to a repository you own.
A private hosted machine starts in about ten seconds.
Fix the auth bug and add a test for it.
plans · executes · commitson its own machine · printed on the answer
In short
Each session gets a machine — ours, started in seconds, or your own, paired free. The agent picks from 300+ models by measured results per job, fans out parallel workers on big builds, and shows the price of every turn.
See it work
Brief it like a contractor: it plans for free, you approve, it executes on its own machine — cost on screen the whole way.
300+ models, each with its price on it.
A spending limit, on from day one.
Side by side
| Question | A coding copilot | Amelia’s Agent |
|---|---|---|
| Shape | Assists you inside an editor or terminal | Executes the whole job on a machine it brings |
| Assumes | You write code | You can describe what you want |
| The machine | Yours, set up by you | Included — or pair your own, free |
| Models | The vendor’s | 300+, each priced, chosen by measured results |
| Publishing | Yours to arrange | One click, certificate included |
Named head-to-heads, scored from their own published pages: Claude Code, Cursor, GitHub Copilot, ChatGPT — or the referee’s Claude Code vs Cursor.
Built like a tool, not a toy
The parts that matter when the work is real.
Models chosen by evidence
300+ models with live prices. Recommendations come from measured builds — which model finished which job, at what cost — not from leaderboards.
Parallel where it pays
Big builds fan out to parallel workers with a lead agent coordinating. Small jobs stay single-agent, because burning tokens on ceremony is not engineering.
Git-first custody
Every turn commits. The repo is yours from the first sentence, mirrorable to GitHub, cloneable from anywhere.
Developers get the deeper page: your repo, your model, your machine — every project is standard git, mirrorable to GitHub. Everyone else never needs to know this page exists — describe a thing and watch it get built.
300+models, each priced per turn
~10sto start a private hosted machine
Real output, not mockups
Correct, not just fast
A calculator that gets the arithmetic right.
“Build a moving-quote calculator. Price is $150 per bedroom plus $2 per mile, and 20% more if packing is ticked. Have it load already set to 3 bedrooms, 40 miles, packing ticked.”
2 turns89s
$0.0047
See all 53 builds →
Give it something hard.
The estimate shows before you sign in — and the first build is free to try.
Questions
Asked, answered
What is an AI coding agent?
Software that takes a goal in plain language, then plans, writes, runs and corrects real code on a real machine until the goal is met — as opposed to a chat model that only suggests code for you to run yourself.
Which models can the agent use?
Over 300 through one account, priced per turn before you spend — or bring your own keys. We publish measured recommendations per kind of job from real builds rather than benchmarks.
Can it work on my existing repository?
Yes — connect GitHub and import it, or run the agent on your own machine so the code never leaves it. It reads the project first and reports before it rewrites.
Does it run on my hardware or yours?
Either. A private hosted machine starts in about ten seconds, or pair your own Mac, Windows or Linux box — free on your own hardware, with real phones plugged in for mobile testing.