ConchClient sign-in

A private trading desk, run by an AI team and supervised by people.

Conch runs perpetual futures trading for one private client, from the client's own account. A team of AI employees does the daily work of a small trading firm. People supervise, review and approve.

Conch is a private desk. It is not open to the public, does not accept investors and is not an offer of any service.

The team

Each member is an AI, not a person. Each has one job, a short list of tools, and a spending limit. None of them can size, price or sign an order.

  • Head of desk · AI

    Writes the plan for the day and a memo with a recommendation for every decision people have to make.

  • Market analyst · AI

    Writes the daily brief, and opens it by scoring the previous brief against what actually happened.

  • Screener and pattern scout · AI

    Turns a plain-language idea into a fixed, repeatable screen and raises alerts with context.

  • Quant researcher · AI

    Turns ideas into strategy specifications and tests them, with trading costs included.

  • Risk officer · AI

    Explains every refusal in plain words and watches exposure. It can stop the desk; it cannot loosen a limit.

  • Execution trader · AI

    Explains how an approved order will be worked and reviews how it filled. It signs nothing.

  • Operations · AI

    Checks the desk's books against the exchange and keeps the incident log.

  • Reporter · AI

    Writes the client's reports from the books: what was done, why, and what waits for approval.

How a trade happens

A language model never decides the size or price of an order, and never signs one.

  1. Step one

    Propose

    An AI employee writes a proposal in a fixed format, with its reasons.

  2. Step two

    Compute

    A quant engine computes the signal from completed market data and fixed rules. The same inputs always give the same answer.

  3. Step three

    Check

    A risk engine approves or refuses the exact order against limits written in code. Every refusal is recorded with its reason.

  4. Step four

    Execute

    An execution service builds and signs only the order that was approved, and the books are checked against the exchange.

The safety model

The client keeps control
The account and the funds belong to the client. The desk never holds the client's own key.
A key that cannot withdraw
The desk signs with a separate trading key that the client approves and can remove at any time. That key cannot withdraw funds.
Limits in code
Which assets, how much leverage, how large a position and how much loss in a day are enforced by software. No AI employee can change them.
A kill switch
One switch stops everything. Turning it back on takes a named person's recorded decision.
Earned in stages
The desk moves from reading, to testing, to paper trading, to small real trades, one stage at a time. Each step needs a person's approval, kept with the evidence they saw.
Everything is recorded
Every proposal, check, order and approval is written to a record that cannot be edited.

What stays private

Positions, results, strategies and the client's identity are never published. Nothing on this page comes from the desk's systems.