The network
Nine months, six products, one token
The honest version, because a timeline is the easiest claim in the world to check. First commit October 2025. Everything below was built since.
How it was built
Market data
The first service: real-time feeds, the thing every other product would need.
Exchange, mobile, custody
The exchange backend, the Flutter app, the admin console and the custody layer, within five weeks of each other.
Midium
The prediction game. Live in both app stores today.
Predikon
A prediction exchange. Live, unattended, and trading demo funds while custody and licensing are settled.
Chain and platform
The Solidity layer for Midas Chain, and the compensation platform the partner programme runs on.
Roughly nine months, and three of the products are under two months old. We would rather state that than imply a longer history that a public commit log contradicts in ten seconds.
Where it goes
The direction, stated as a sequence rather than a set of promises. Each step depends on the one before it.
An exchange that works
Spot, futures, margin, options, P2P, copy trading, savings and fiat rails. Built and live.
Products people use daily
Midium for play, Predikon for forecasting. Both live, both bringing people in who did not arrive looking for an exchange.
Predikon, decentralised
Moving prediction markets from our custody to a non-custodial model. A different product, not a port — and it is a decision taken, not a thing shipped.
A chain of our own
Midas Chain: the venue MIDS is native to, and the listing bond that gives the token its one designed demand sink.
A card
Spending balances directly. No specification, no issuer, no sponsor bank — this is intent, and card programmes take years, not quarters.
What we actually solve
Three things, and we will not claim a fourth.
The exchange is the distribution
Most token projects launch a token and then hunt for somewhere to trade it. MIDS launches into a venue that already has order books, custody, compliance and users — and D20 requires it to be listed with real depth before a single package is sold.
Products that are not a trading screen
A game and a prediction market bring in people who would never open a futures terminal. That is a distribution advantage a pure exchange does not have, and it is why the token touches three products rather than one.
We publish what is not working
The whitepaper states there is no material demand sink yet, that the contracts are unaudited, and that the token is not listed. An issuer that only publishes good news has told you what kind of issuer it is.
On “AI-powered”
Worth being precise about, because the phrase is used loosely across this industry and the precise version is more interesting. There are three distinct things here — two uses of a language model and one data feed — and none of them trades.
Turning a goal into a plan
- You describe what you are saving for in your own words, out loud or typed, and the model turns it into a plan — what it costs, by when, and what you would set aside each month.
- It is muzzled: the model supplies words, the server owns every number. A figure the server did not itself calculate is rejected before it reaches you.
- Every projection carries its worst case. The system cannot return one without it.
Answering questions
- An assistant that explains concepts and your own figures, and refuses to recommend. Ask it what something means and it answers; ask it what to buy, or where a price is going, and it declines.
- Three independent guards enforce that: a pattern filter on the question before any model runs, a model instructed to refuse advice-shaped questions outright, and the same numeric muzzle on everything it returns. Any one failing still leaves the others.
- Single-turn and stateless. There is no conversation history, which means there is no history to poison and no multi-turn jailbreak surface.
Reading the room
- A daily market deck reads social interaction volume and a sentiment score from a third-party feed, and shows where attention has moved away from price.
- It is a data subscription, not a model. Nothing is inferred — the numbers are counts and a published score.
- Information only, and wired to nothing: no basket, no allocation, no execution path. It writes cards to a deck. Without the feed the deck runs on our own market data and still works.
What none of them do
None of them makes or executes a trading decision. Nothing here picks an asset, sizes a position or times an entry, and no signal is connected to anything that can place an order. The trading bots are rules, not models. Anyone telling you an AI trades your money here is wrong, and we would like to know where they heard it.