Private ยท trading and risk management

desk

A trading and risk desk with an AI analyst on staff.

Desk manages an Interactive Brokers book through the trading day: stop losses and hedges set by rule, a written brief before the open and after the close, AI-augmented research and decisions with dated evidence, and the portfolio numbers that matter, from exposure to time-weighted performance.

The Desk console showing the Portfolio view: an account panel, a portfolio performance chart with drawdowns and transfers, a holdings table with intraday charts, a market ribbon, and the Ask Desk panel on the right
Portfolio, in simulation mode: a synthetic paper book, live-style quotes, performance from synthetic statements, Ask Desk on the right. Every screenshot on this page comes from that mode.

What it does

Rules first, then briefs, then AI where judgment is needed.

Automatic stop losses

Canary's protection rules place broker-side trailing stops, keep exit watches and hold hedges per position, with no AI in the loop. Desk shows each one on the holding line and under Risk.

Daily briefs

A written brief at 08:30 Berlin, half an hour before the US open and at 18:00: positions, risk, the calendar, what changed, what needs you.

AI-augmented investing

Scheduled reviews at the open, before the close and after it. An analyst model proposes adds, holds, trims and exits with a reason, sources, what would prove it wrong and an expiry. A second model reviews before anything reaches you.

Ask Desk

A durable conversation with the book: which positions need attention, exposure and regime, the next session. Evidence is gathered first. Missing evidence means no recommendation.

Portfolio management

Exposure, margin and Greeks, sector and asset mix, reconciliation against statements, and time-weighted performance against SPY, net of transfers.

On the market's clock

07:00 Berlin, an hour before each open, at the open, before the close, after it. Session times come from Canary, early closes included, so a holiday never runs a review.

How it is built

Desk is mostly configuration. Canary owns the broker: session, prices, risk, calendars, stops and hedges, reconciliation. Torok, the agent harness, owns the work: inbox, schedules, bounded episodes, memory, recovery. HyperServe serves the console on loopback. Desk adds the instructions, nineteen read-only Canary tools, the observers and the shape of a decision. Models come from OpenAI. One analyses, another reviews and researches.

System diagram of Desk: the owner, passkey, notifications and macOS on the left; the Desk process with HyperServe, the Torok resident service and Desk's capabilities in the middle; the Canary daemon and its three access paths beside it; TWS, IBKR statements, market sources, official sources and the OpenAI API on the right
Every component Desk runs with. Green is the broker path, blue is observed data, amber is a paid model call.

Screens

Five views, one masthead, one quote ribbon, Ask Desk on the right. Every figure carries its source and its age.

What it is not

  • Not a product. No download, no repository.
  • No claim of edge. Correct operation and investment results are measured separately. Building it proves neither.
  • One machine. A user LaunchAgent, running while the Mac is awake and logged in.
  • Simulation in the pictures. Real screens look the same with real numbers, which is why they are not here.

Contact

Desk stays private, but the ideas travel: rules before models, evidence contracts, a reviewer that is not the analyst. If you are building something like it, get in touch.