Pricing
The assistant, the screener and the backtester cost real money to run, and the honest way to sell them is to say which part costs what. Start free with no card and no API key. Pay when the limits start to matter. Trading with real money is a separate decision, and it is ours as well as yours.
Every tier runs the same platform: the same screener, the same backtester, the same risk gate between the model and an order. What changes is how deep the price history goes, how often a saved screen runs, how much you can ask the assistant, and whether real orders can leave the building. Only the free tier can be started today — the paid ones are built and not yet purchasable, and you can register your choice from inside the console so it is waiting when they open.
€0
No card. No API key. No end date.
Twenty conversations a month on our model, three saved screens that run once a day, six months of history in the backtester and one paper account. Enough to run the whole system and decide whether it is any good — not enough to run a trading week on.
€29
Per month. The model is included.
For the reader who has stopped evaluating and started using it. Six hundred conversations, twenty-five screens running four times an hour, and the full two years of history the backtester can reach.
Not yet purchasable. Start on the free tier and choose this one in the console — we will bring it up first for the people who asked.
€59
Per month. For people in the market daily.
Screens every five minutes, a hundred of them, and the strategy search — the part that spends real CPU proposing and backtesting candidates on its own. Five paper accounts, so a thesis can be tested in parallel rather than in sequence.
Not yet purchasable. Start on the free tier and choose this one in the console — we will bring it up first for the people who asked.
€99
Per month, and not self-serve.
Everything in Pro, plus a connected exchange and real orders. Your venue keys sit in a process that serves you and nobody else — that is infrastructure per customer, and it is the reason this tier costs more than the compute inside it.
Enabled by approval, not by a checkbox. Ask from inside the console once you have run the system on paper; we turn trading on for your account after a conversation. Nobody gets a key field to fill in on their first day.
| Included | Free | Console | Pro | Live |
|---|---|---|---|---|
| Conversations per month | 20 | 600 | 2,000 | 2,000 |
| Bring your own model key | — | Optional | Optional | Optional |
| Saved screens | 3, run daily | 25, every 15 min | 100, every 5 min | 100, every 5 min |
| Backtests per month | 20 | 300 | 1,500 | 1,500 |
| Price history the backtester reaches | 6 months | 2 years | 2 years | 2 years |
| Paper accounts | 1 | 3 | 5 | 5 |
| Strategy search runs | — | — | 10 per month | 10 per month |
| Connect an exchange | — | — | — | By approval |
| Live orders, through the risk gate | — | — | — | Yes |
One conversation is one thread with the assistant — a question, everything it looks up to answer it, and the follow-ups you ask in the same session. Asking what changed in your positions costs one. Asking it to research a thesis across six instruments also costs one, and costs us a great deal more; the console shows you what a thread has consumed while you are inside it. We quote conversations because that is the unit you think in, and we meter the underlying cost because that is the unit we are billed in. If the two ever diverge badly, we would rather change the number on this page than quietly put a smaller model behind it.
An earlier draft of this page made the free tier bring-your-own-key: unlimited conversations, paid for by you, at your own provider. It reads generous and it is the wrong shape. Nearly every product that ships bring-your-own-key is self-hosted or built for engineers, because obtaining an API key is a step most people never take — and a free tier that opens with "first, create an account at OpenAI" is a front door nobody walks through. So the free tier runs on our model with twenty conversations, it costs us money, and that is simply what a funnel costs. Bringing your own key stays available on the paid tiers, where it is a real convenience for someone who already holds provider credit and wants to choose the model per conversation.
The language model is the visible expense and it is not the largest one. Market data arrives continuously whether or not you are logged in. A saved screen running every five minutes is 288 evaluations a day against live candles. A backtest across two years is a compute job, the strategy search is many of them in a row, and the sandbox that runs your own strategy code is a container we start for you. That is why the tiers differ on history depth, screen frequency and search runs rather than on which buttons are enabled — those four things are what we are actually billed for, and pretending otherwise would produce limits that make no sense to anyone.
It is not scarcity marketing. A process holding one customer’s exchange keys is a different risk from a process holding fifty, so the money path is isolated per paying account rather than shared — real infrastructure per customer, and the reason the tier is priced above the compute it uses. The gate is useful in the other direction too: a stranger cannot register at breakfast and hold a trading-enabled API key by lunch. New accounts get the assistant and paper trading; connecting an exchange is a conversation first, and we will ask you to create a key that cannot withdraw.
The feature stops until the month rolls over and says which number it hit. Nothing degrades silently, nothing is billed as an overage you discover later, and no order behaves differently because a quota is low. If you hit the same ceiling repeatedly, the answer is usually the next tier and occasionally it is that our number is wrong — tell us which one and we will look at it.
No. It has no end date and no card. It has limits, they are printed above, and they are the ones that cost us money to lift.
Thirty times the conversations, screens that run four times an hour instead of once a day, fifteen times the backtests, and the full two-year history. If you look at the market once a week, the free tier is genuinely the right choice and we would rather you stayed on it until it stops being.
Screens every five minutes instead of every fifteen, four times as many of them, and the strategy search — the assistant proposing and backtesting candidates by itself rather than only answering what you ask. It is the tier for someone whose day already has the market in it.
On any paid tier, at launch — OpenAI, Anthropic or OpenRouter. It does not lower the subscription, because the subscription is for the platform rather than for the model; it is worth doing if you want a specific model, or already hold credit you would rather spend. It is not offered on the free tier, where the point is that you need nothing to start.
Ask from inside the console, once you have run the system on paper. We enable it for your account by hand. Expect to be asked what you intend to trade and at what size, and expect us to insist on an exchange key that cannot withdraw.
Any time, from the console, effective at the end of the period you have paid for. Your history, screens and strategies stay readable on the free tier — cancelling is a downgrade, not a deletion.
No card, no API key, no exchange connected. Run the screener, break a strategy in the backtester, and decide about a subscription after the platform has been wrong in front of you at least once.