Build your own
Your own AI trading desk. Built to tell you the truth.
The complete blueprint behind my autonomous multi-agent paper-trading operation: specialist desks, a manager that audits them every morning, and a validation gauntlet a bad idea cannot talk its way past.
What this is
Not signals. A research organization.
Most AI-trading products sell you signals: someone else's black box, and you get the outputs. This kit teaches you to build the organization instead — agents that trade simulated capital under hard risk rules, a manager that audits them daily and hunts for new edges on weekends, and a testing process with gates you register before you see the results.
Everything in it was extracted from a real, running system — not a thought experiment. You get the architecture as it actually stands, the lessons in the order we learned them, and the failures that shaped every guardrail.
The honest scoreboard
Our gauntlet rejected 13 ideas out of 13. That's the product.
In its first weeks, our validation engine tested thirteen strategy ideas and honestly rejected all thirteen. Our research desk ran another forty-two hypotheses through its own seven-gate process — also zero survivors.
We lead with that number because it is the feature. A system that only shows you winners is a system that cannot say no — which makes its yes worthless. This one says no constantly, in writing, with the evidence attached, and raises its own statistical bar with every failed trial.
A system that can tell you no is the only kind whose yes will ever mean anything. If you want software that flatters you, this kit will frustrate you. If you want a process you can trust, this is it.
What you receive
Four folders. Everything you need.
The kit is organized as a numbered path — read it in order, or go straight to the build prompt and let the lessons catch up with you.
The Architecture
How the whole system is structured, what every agent on the roster does, and the master build prompt — a single document you paste into Claude Code that constructs your starter system in phases and interviews you for every decision that is genuinely yours.
The Playbook
Ten lessons in the order we actually lived them: the philosophy, the guardrails, your first desk, keeping a fleet running unattended, the manager, the validation gauntlet, the research pipeline, and what broke along the way. The failures are included on purpose — they are where the value is.
The Knowledge Compendium
The research library: fifteen dossiers on trading strategy families — trend following, mean reversion, pairs trading, factor investing, Wyckoff, and more — each with its evidence explicitly sorted into what is established and what is speculation. Plus the reusable protocols for producing new dossiers of your own.
The Change Log
Every meaningful change our system made to itself and the reasoning behind it, plus the discipline for keeping your own. This audit trail is what makes an autonomous system trustworthy rather than mysterious.
What you build with it
A fund that runs itself — and answers to you.
A manager that audits everything
The Hedge Fund Manager agent inspects every desk each morning, researches new ideas on weekends, and reports to you by Telegram. It has authority over analysis and zero authority over capital — it is physically unable to trade, edit a desk, or touch a credential.
Specialist desks under hard rails
Each trading agent runs simulated capital inside per-trade risk caps, exposure caps, and a daily loss halt. Every decision is journaled. Every desk reviews its own week in writing.
Approval-gated evolution
Desks may propose changes to their own strategy — and nothing applies until the proposal passes a backtest and you approve it in writing. The system improves; it never drifts.
A gauntlet that can say no
New strategy ideas face pre-registered pass/fail gates, a once-only holdout test, a statistical bar that rises with every trial, and an adversarial reviewer with veto power. A bad idea cannot talk its way past it.
Paper trading by construction
The simulated-trading endpoint is a hard-coded constant guarded by a startup check — the process refuses to run if it is edited. Going live is a human decision this kit deliberately does not automate.
You as the PM
You approve proposals from your phone, drop in ideas, and read the daily digests. Final say on everything stays with you — the machine does the work, you run the fund.
Requirements
What you need before you start.
- Claude Code (the desktop app) with an active subscription
- Python 3.11 or newer installed
- A free Alpaca paper-trading account — simulated money, no real funds
- A free Telegram account — your remote control and daily reports
- Optional: Notion for structured logging, and free-tier market-data keys
No coding required. The agents do the engineering; the build prompt directs them, and the Playbook makes you a competent owner of the result.
Read this before buying
What this is not.
Not live trading.
Every agent in this architecture hard-codes the simulated-trading endpoint and refuses to start if it's changed. Going live is a human decision the kit deliberately does not automate — the final lesson covers the checklist we would demand of ourselves first.
Not a promise of profits.
Our own honest scoreboard is printed above and in the Playbook. What the kit guarantees is a process that will not lie to you about what works — not a particular result.
Not financial advice.
This is educational software and educational writing about architecture and research process. Nothing in it is a recommendation to buy or sell anything.
Pricing
Two ways to get it.
Buy the blueprint and build at your own pace, or have me alongside you from the first prompt to a running fleet.
The Kit
$2,000
Self-paced. The complete blueprint, ready to build from on day one.
- The full architecture and agent roster
- The master build prompt for Claude Code
- The ten-lesson Playbook
- Fifteen research dossiers plus the research protocols
- The change-log discipline and templates
- Single-user license — build it, run it, modify it freely
Kit + Build With Me
$5,000
Everything in the Kit, built together — with support that continues after launch.
- Everything in the Kit
- Working sessions with me to stand up the manager and your starter desks
- Your first audit cycle and first gauntlet run, reviewed together
- Help extending to specialist desks when you're ready
- Ongoing support after launch
FAQ
Common questions.
Do I need to be able to code?
No. You paste the build prompt into Claude Code and the agents do the engineering. Your job is to read, decide, and approve — the Playbook teaches you to do that well.
Will the system make me money?
The honest answer: it trades simulated money, and our own gauntlet has so far rejected every strategy it tested. What you are buying is the research organization and the discipline — a machine that finds out what's true. Anyone who promises you more than that is selling the thing this kit exists to protect you from.
Does it ever touch real money?
No. Paper trading is enforced in the structure of the code, not by a policy. If you ever chose to go live, that would be your own separate decision — the kit deliberately does not automate it.
What does it cost to run?
Your existing Claude Code subscription plus free-tier accounts: Alpaca paper trading and Telegram are free, and the optional market-data keys have free tiers. There is no required additional spend.
How is this different from the Operator's Stack?
The Operator's Stack teaches you to run a small AI team in any domain. The Fund Kit is the deep end of one domain: a complete, self-auditing research organization with validation discipline. They share the same operating philosophy — many buyers of one end up wanting the other.
Build the fund that can't lie to you.
The blueprint, the lessons, the library, and the audit trail — extracted from a system that runs every day.
See pricing