conw.ai

Usefulchatsteach it.Seewhatstuck.

Conw is independently served. Teach it something and it works from the next reply — every reply that used a memory shows you which one, with a one-tap Forget.

No card required 30-second setup Free to use
Verified learningIndependently servedLocal MLX inferenceLearns from useful chatsLiving-room iMacFree to useVerified learningIndependently servedLocal MLX inferenceLearns from useful chatsLiving-room iMacFree to use

How it learns

You teach. It filters. It proves what stuck.

A guarded learning loop that improves from useful signals without letting one bad chat rewrite the system.

01

You teach

Correct a reply, rate it, or explain a new word. Confirmed meanings enter your private memory immediately so Conw can use them in the same conversation.

02

It filters

Sensitive, unsafe, badly formatted, and down-rated replies stay out. Only short, useful examples can enter the reviewed learning queue.

03

It proves it

Background updates wait for enough clean examples and an idle server. A candidate must pass checks before it can replace anything live.

Confirmed teaching works immediately. Weight updates never block your reply.

The night shift

When the iMac is quiet, it reviews.

Eligible examples wait in a guarded queue. Training starts only after there is enough clean material and serving is idle. The result is a candidate, not an automatic live update.

Signal

Only safe, useful, short replies survive. Down-rated and contaminated replies stay out.

Schedule

Background work waits for the iMac to be idle so live chat keeps priority.

Promotion

A candidate must load and pass its checks before anything live can change.

Ω

guarded queue

live chat first · learning second

Principles

Three things we can say with a straight face.

01.

Most AI products hide the learning loop. Conw shows it: what got taught, what got rejected and why, live at conw.ai/learning.

Self-taught

It learns from the people who use it.

Teach it a preference, a fact, or a word and it's written to your private memory immediately — the next reply that uses it shows you exactly which memory, with a one-tap Forget. A taught word can go further: once several independent people teach the same thing days apart, it's checked, reworded so it's never anyone's literal sentence, and quizzed before it can help every user.

  • Confirmed memories work immediately, visible in every reply that uses one
  • Corroborated word meanings can help every user, never your personal facts
  • Unsafe, invented, or poisoned lessons stay out — and rejections are public too
0checks before a taught word becomes shared knowledge

02.

Conw is a workshop, not a monument: the serving model can change while memory, safety, verification, and user ownership stay intact.

Independent by design

Our serving stack. Our hardware. A replaceable checkpoint.

Conway-Retrain 12B is now live, retrained on top of a Gemma 4 base and served through MLX on our own machine, not an external answer API. The earlier 188M Conway-Omega checkpoint is deprecated in-product, but stays published, open-source, on Hugging Face. The checkpoint stays behind an interface, so memory and safety rules survive future model changes.

  • Inference is served by Conw, not forwarded to another model API
  • Checkpoint-specific thresholds live in calibration, not product code
  • The learning loop survives future checkpoint replacements
0local serving host for the current product

03.

The greenest token is the one we never need to generate. Efficiency starts with useful answers, bounded output, and training that never starves the person waiting.

Light by design

Efficiency we can measure, not invent.

Conway-Retrain 12B currently serves from one 16GB iMac. Short-answer token ceilings avoid waste, repetition guards stop runaway output, and background learning yields whenever chat traffic needs the machine. We will not publish a carbon number until we can meter it properly.

  • One 16GB serving machine instead of a GPU cluster
  • Response budgets stop needless token generation
  • Background learning defers when inference is busy
0GBmemory budget for serving and learning

Platform API

Build on Conway-Retrain from your own code.

Pay-as-you-go access to Conway-Retrain 12B and Conway-Omega 188M, live at api.conw.ai — the same model behind the chat product, reachable from any OpenAI SDK.

View the API
  • Drop-in OpenAI-compatible endpoints — swap base_url and go.
  • £1.50 per 1,000,000 tokens, input and output, either model.
  • Named keys, usage graphs, and a £5.00 minimum top-up.

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, and you can watch it happen. Teach it a preference, a fact, or a word and it's saved to your private memory immediately — the next reply that uses it tells you exactly which memory it drew on, with a one-tap Forget button. A small number of taught words can go on to become shared knowledge every user's assistant can draw on, but only once several independent people teach the same thing days apart, and only after it's reworded so it's never anyone's literal sentence. Every rejection is logged too — see the whole thing, successes and failures, at conw.ai/learning.

Get started

Meet it today.
Teach it something useful.

Conw is free to use. Confirmed teaching works immediately, and safe, useful replies can become reviewed learning candidates.

No card. No spam. Free to use.