Not a wrapper. An orchestration layer.
Anyone can put a chat box in front of a language model. What a broker is actually buying here is everything between the question and the answer: which model runs, what it is given, which specialists are called, and what is checked before a figure reaches a client.
In the client's own words
Prices, news, on-chain, flows
Positions and history, if you allow it
At the edge, not at the centre. A model is a component here, and components get replaced when better ones ship.
Four things a wrapper cannot do.
A wrapper sends your question to one model
Zeig picks the model per step. Reading a chart, reasoning over a position and writing a summary are three different jobs, and the best model for each is not the same model, nor the same one it was six months ago.
A wrapper answers from what the model already knows
Every answer here is grounded first: live prices, the news record, on-chain and flow data, and where you allow it, the client's own positions. The model is given the facts rather than asked to remember them.
A wrapper gives you one voice
Zeig runs specialists. A question about a chart, about a week of behaviour and about an account's exposure go to different agents, and the layer decides which of them the question actually needed.
A wrapper hands back whatever came out
Figures are held to the data they came from before the answer is returned. A number that cannot be traced does not ship, which is the difference between a research tool and a confident guess.
What this buys a broker.
You are not buying a model
Models change every few months. Because they sit at the edge of this architecture rather than at its centre, a better one is an upgrade rather than a rebuild, and that work is ours rather than yours.
No single vendor holds your product
The layer speaks to several providers. If one changes its pricing, its terms or its availability, your client-facing product does not change with it.
It is built for markets, not adapted to them
Timeframes, levels, exposure, drawdown and position sizing are in the domain logic rather than in a prompt. That is three years of work that does not start again with each model release.
Three years, by people who have traded.
Zeig has been in engineering for three years, by a team of financial and technology specialists whose backgrounds include JP Morgan and Goldman Sachs. That matters less as a credential than as an explanation of why the domain logic is where it is: the people who decided what a drawdown figure has to survive before it is shown to a client had spent years being accountable for figures like it.
The platform runs on the infrastructure the market already uses. Charts are TradingView. Execution is designed against MetaTrader and cTrader. Zeig is the intelligence between them, not a replacement for either.