Why I built this
I lost money trading. So I built the thing I needed.
Four minutes: the grid bot that blew up, why a reasoning model changes the problem, and exactly how the three agents work.
AI agent trading crypto for you
Soulquant runs your strategy on your own exchange account — not as a bot firing on candles, but as a small team of AI agents that reason, check each other, and learn from what they got wrong.
On the record
One of our own agents, running on a real account. Every wakeup, every decision to pass, every trade — published as it happens, with the reasoning that produced it. No signup to read it.
Most wakeups end in nothing to do here. That is not the agent being idle — it is the agent being disciplined, and it is the number most systems hide.
The system
Judgment, execution and learning never share a hand. That separation is the whole design — it is what stops a confident model from talking itself into a bad trade.
Indicators, volume profile, the regime, the levels that matter. It defines which events should make it reconsider, and wakes itself when one happens. It signals a trade — side, entry, stop, target, confidence — and never an amount.
It reads your live balance, sizes against your risk profile, and rejects deterministically before any order if a gate fails. Nothing reaches your exchange without it. The agent that finds a trade is not the one allowed to place it.
On a slow cadence it reads the whole logbook back — with the same market tools the other two had at the time — and writes what the period actually taught. It never places a trade.
Your doctrine
The engine defines how an agent behaves — not what it trades. That is deliberate: it means the strategy on top can be almost anything, as long as the market data it needs exists. Start from a template, or describe your own in plain language and the wizard drafts it.
Describe how you trade in a conversation. The wizard drafts it, checks it for consistency, and won't finalise until it holds together.
By construction
No system is right about a market. The question is what happens when it isn't — and that is a property of the architecture, not a promise in a headline.
Trade-only API keys on the exchange you already use, encrypted at rest and decrypted only at run time. Never withdrawal scope, never logged. We are the infrastructure — we never hold, custody, or move your money.
Every agent starts simulated against real market data, and going live is a deliberate, confirmed switch. While it's on paper, no code path can reach a real key — that's enforced, not intended.
The model never picks an amount. Position size, leverage caps and risk gates are deterministic code, and a failing gate rejects the trade before any order exists.
Why I built this
Four minutes: the grid bot that blew up, why a reasoning model changes the problem, and exactly how the three agents work.
Pricing
I want people actually using this, and I want to learn from what happens when they do. So for the first ten: no platform fee, ever. You start with free credit, and after that you only pay for the compute your agents spend thinking — and a patient agent thinks rarely.
$39$0 / month
No platform fee, ever. You start with free credit.
Pay as you go
You pay for the tokens your agents actually use, and nothing while they sleep. Metered per run, visible per run.
Signing up and running on paper is free for everyone — no card required. After the first ten seats, a platform seat is priced from $39/month and locked for life at the rate you joined at.
FAQ
Pick a template, connect your keys, and watch it work on paper for a few days. Then decide. This is early, it will keep getting better, and the people here now see all of it.
Paper mode first · no card required · we never touch your funds · no performance promises, ever