Everychatteaches it.Comebacktomorrow.

Conw is an independent AI built from scratch. Conway-Omega learns from the conversations people have with it — the model you talk to next week is sharper than the one you talk to today.

No card required 30-second setup Free to use

Sharper every dayBuilt from scratch500M parametersLearns from every chatLiving-room iMacFree to useSharper every dayBuilt from scratch500M parametersLearns from every chatLiving-room iMacFree to use

How it learns

You talk. It trains. It comes back sharper.

This is the whole product. Not a feature list — a loop that never stops running.

You talk

Every conversation is a lesson. The questions people ask — and the ways Conw fumbles them — become training signal.

It trains

Conway-Omega is small enough to keep retraining on recent chats. Frontier models freeze their weights for months. Ours never sits still.

It comes back sharper

The next time you open a chat, the model answering you has learned from everything since your last visit. Same name, better brain.

The loop is running while you read this.

The night shift

While you sleep, it studies.

Today's conversations become tonight's homework. Conway-Omega retrains on what people actually asked — so the version that greets you in the morning has already learned from yesterday's mistakes.

Signal

Real conversations become training data — the good answers and the fumbles alike.

Cadence

Weights update continuously. Not yearly. Not "when the next version ships."

Result

Tomorrow you talk to a newer model than you did today. Same name, better brain.

in training

the training loop · always on

Principles

Three things we can say with a straight face.

01.

Most AI products hand you a frozen model and call it a day. Ours is still in school — and you're the teacher.

Self-taught

It learns from the people who use it.

Conway-Omega retrains on real conversations. That means it improves where people actually push it — not where a benchmark says it should. Catch it being dumb today, and you've written part of tomorrow's lesson plan.

Real chats become training signal
Weights update continuously — not once a year
The model you shape is the model you get

0/7in training, not on a shelf

02.

Frontier models are cathedrals: magnificent, and finished. Conw is a workshop — smaller, louder, and open every day.

Small on purpose

500 million parameters. Zero borrowed weights.

Conway-Omega was trained from scratch — tokenizer, data pipeline, weights, all ours. It's a small model, and we won't pretend otherwise: the giants will beat it at trivia. But small is exactly what makes constant retraining affordable. That trade is the product.

Not a wrapper around someone else’s API
Small enough to retrain constantly
Honest about its limits — that’s the deal

0Mparameters, built from scratch

03.

The greenest token is the one a 500M model can generate without waking a warehouse of GPUs. Efficiency isn't our marketing — it's our architecture.

Light by design

Green by arithmetic, not by offsets.

We make exactly one environmental claim we can stand behind: Conway-Omega is a fraction of the size of frontier models, so each answer costs a fraction of the energy. Right now it runs on a small iMac in Thomas Conway's living room. That's the whole claim. No certificates, no invented ledgers.

A tiny model burns tiny energy per answer
Hosted on a small iMac in Thomas Conway's living room
No greenwash — small is the strategy

0×smaller than frontier models

Why we tell the truth

“Most AI marketing is fiction. Ours can't afford to be — you'd notice tomorrow if we lied today.”

A model that improves in public has nowhere to hide.

FAQ

Straight answers.

Anything we missed? [email protected] — a human writes back.

Yes — that's the whole product. Conway-Omega retrains on the conversations people have with it, so it genuinely improves over time. It also means you shouldn't paste anything you wouldn't want a model to learn from.
It's a 500-million-parameter model, and we won't pretend otherwise — the giant models will beat it on breadth and trivia. Its edge is the loop: it keeps training on real conversations, so it gets sharper exactly where people actually use it. If it fumbles your question today, you've just contributed to tomorrow's training run.
We keep this claim modest, because most AI energy claims aren't. Two things are true: Conway-Omega is a fraction of the size of frontier models, so each answer takes a fraction of the energy — and right now it runs on a small iMac in Thomas Conway's living room. That's the entire claim. No offsets, no certificates, no invented ledgers.
No. Conway-Omega was trained from scratch — tokenizer, data pipeline, and weights are all ours. When you talk to Conw, you are talking to Conway-Omega, not a rebadged API.
A small model is cheap to run, and every conversation makes it better — that's the exchange. The Free plan covers casual weekly use; Pro and Max just raise your weekly limits and pay the server bill. Same model, same answers on every tier. No ads, no selling your data.
Conversations are stored so the model can learn from them. We don’t run ads and we don’t sell data. If you want your account and chats removed, email us and we’ll delete them.

Get started

Meet it today.
It'll be sharper tomorrow.

Conw is free to use, and every conversation teaches it something. The model you shape is the model you get.

No card. No spam. Free to use.