An assistant thathas read your book
Not a chat window bolted onto a chart. It reads live market data, your open positions and the decisions behind them, and it works in sessions that remember what you concluded last week — while every action that could cost money goes through the same gate as everything else.
Proposed Thesis check requested — funding turned positive, trend intact
- Position readyour account only
- Market dataclosed candles · 4h
- Action takennone — this is a note
- Order draftedrequires your confirmation
What it is actually able to do
Read the market, properly
Candles at every interval, indicators computed the same way the screener and the backtester compute them, order flow, funding and open interest. It quotes the numbers it used, so a claim it makes is one you can check rather than one you have to accept.
Remember across sessions
Conclusions worth keeping become notes attached to your account — the thesis behind a position, what you decided about an instrument, the rule you set for yourself in March. A new conversation starts with those in front of it, which is the difference between an assistant and a search box.
Watch a position and tell you when the thesis breaks
Ask it to follow a position and it re-checks on a schedule, writing what it finds back into the same conversation and to Telegram if you want it there. When it thinks something has changed enough to act, it proposes — and the proposal waits for you.
Say what it does not know
Where a number is missing it says so instead of substituting one. An assistant that fills a gap with a plausible figure is worse than no assistant on a page where the figures are your money.
And what it is not able to do
The list matters more than the capabilities, because this is the part you are being asked to take on faith everywhere else.
- Place an order that has not passed your risk limits. There is no tool that submits directly.
- Change those limits. It can show you where they are; it cannot write them.
- Move money anywhere, including between your own accounts.
- See another customer’s account, positions or conversation.
- Quietly act on a schedule it set itself. A recurring watch is one you created, and you can see and stop it.
Questions people ask about the assistant
Which model is behind it?
A ladder of them, chosen per task: a small fast one for routine reads, a stronger one where the reasoning is worth paying for. The boundary does not depend on which one answers — a better model gets no additional privileges.
Does it work while I am asleep?
Watchers and screeners do. They report back and, at most, propose. If you have promoted a strategy to trade automatically, that strategy trades under your limits and its own rules — the assistant is not the thing that decides in the night.
What does it do with my conversations?
They are yours, kept with your account so the assistant can remember them. They are not another customer’s context, and they are not training material.
Can it be talked into doing something it should not?
It can be talked into wanting to. The design assumption is that a model will occasionally be persuaded, which is why the things that matter are not requests it makes politely but calls that do not exist and signatures it cannot forge.
Ask it something hard, on paper
A paper account gets the whole assistant, live data, screeners and backtests. Connect an exchange later, or never.