Private beta · paper trading
An AI-native trading desk. You set the line.
A team of Claude agents reads the market, writes the thesis, proposes the trade, argues the other side, and learns from every decision it makes. One deterministic program, the Executor, is the only thing that can touch the broker, inside limits you set. Nothing moves without your line.
- Paper money by default
- Never moves money
- Every decision journaled
Built to learn
It learns. Every single day.
Every decision the desk makes is written down and scored against what the market actually did: the good theses and the bad, the wins and the losses. What holds up becomes memory, what does not is retired, and tomorrow’s brief starts from what today taught. It is built so the desk you sign in to next month knows more than the one you meet today.
The loop, clockwise: Decide, The market answers, Scored by code, Kept or retired, Remembered, Tomorrow’s brief, and back to Decide. Three laps, one for each trading day; the stream comes back thicker each time.
Each lap is one trading day, and the stream comes back thicker than it left.
What it is
A desk that never sleeps.
Bellwether reads the market the way a professional desk does: the overnight news, the filings, the price levels and the order flow. It forms a view with conviction, writes down what would prove it wrong, and argues with itself before anything reaches you. When the facts change it changes its mind, and says so. It acts only inside the line you draw. Underneath, time‑tested, published market signals are fused with our own desk intelligence: the process is what we build, not a secret indicator.
Paper money by default
Every book starts on a paper account with real prices and pretend money, and stays there until weeks of evidence say otherwise.
Never moves money
At your broker it can read the account and place or cancel orders, nothing else: no deposits, no withdrawals, no bank link. You fund the account yourself, at the broker.
Every decision journaled
Each brief, proposal, objection and order is written down with who, when and why, then scored against what the market did.
The rhythm
How a day runs
The same beats every trading day, Chicago time. You can watch all of it or none of it; the line you set holds either way.
07:45
Morning brief
Researcher
Overnight news, filings and levels for the stocks on your watchlist and in your book, read and written up as a short brief with its sources.
08:00
Proposals
Portfolio Manager and Critic
A thesis with the evidence it rests on, a size, a stop and the price that proves it wrong. The Critic attaches its objections.
Before the open
Your line
You
Anything above the approval line you chose waits for you. Below it, your autonomy setting decides. Nothing else moves.
08:30
The open
Executor
Waits out the first minutes, re-checks price, size and every limit at the moment of sending, then places one day limit order and manages its stop.
09:00 to 14:30
Hourly checks
Portfolio Manager
Levels reached, theses drifting, stops to tighten. Most hours it does nothing, and writes down why.
14:48
Before the close
Portfolio Manager and Executor
A last pass twelve minutes before the bell. Anything that should not be held overnight is proposed for sale, inside the same line and the same checks.
15:15
Review and lessons
Critic and Portfolio Manager
The session is journaled and every proposal is scored against what actually happened. The findings are on the desk tomorrow morning.
Overnight
While you sleep
The whole desk
Lessons are kept or retired, calibration is re‑checked, and tomorrow’s brief starts from what today taught. The line still holds.
The team
Meet the desk
Five Claude agents with narrow jobs, one more in testing, and one plain program that holds the keys. They are kept deliberately apart: the agents propose, code disposes.
Researcher
Reads the news, the filings and the levels before the open and at midday, and writes a short brief with its sources.
Portfolio Manager
Turns the brief into proposals: a thesis, a size, a stop and the price that would prove it wrong. It can only ask; it never sends an order.
Critic
Argues the other side of every proposal and every session. Its objections are answered in writing, not waved away.
Quant
Tests the published signals against your book’s own history after the close and recommends bounded changes with the evidence attached. Nothing it suggests applies silently.
Consult
Answers your questions about the book in plain English. It has no way to trade.
Trader
In testingSwitched off for now
Built to work out how an approved idea gets placed: the working price, the timing, when to give up. It stays switched off until the prices it gets beat the plain rules that place orders today.
Executor
Plain code, no AI
The only component with the broker keys, and there is no model inside it. It runs every order through the same checks, holds anything over your line for you, and cannot be talked out of a limit.
Second opinion
A desk that argues with itself.
Every proposal meets the Critic, a second agent whose only job is the other side, and its objections are answered in writing or the trade changes. This one is from a recorded paper session in which the Critic’s verdicts were written down but not yet applied: it said shrink the trade, the order went in as proposed, and ten days later the Critic had been right. That score is what the desk learns from.
Premarket proposals
Buy the dip in a payments stock at its 50‑day average.
- Strongest twelve-month momentum among the stocks the desk follows
- Still above its 200‑day average
- A two‑day momentum gauge says oversold
Stop a little under one ordinary day’s move below the entry
- The stop sits inside one ordinary day’s range. Yesterday alone moved further than that.
- Oversold on the two‑day gauge only. The two‑week gauge is neutral, and the slide so far is 3.3%: one average day for this name.
The order went in as proposed and filled. The Critic’s verdict sat beside it in the record, not applied.
Ten days later the stock was about 5% lower. The journal scored the Critic’s objection as right, and that lesson goes into tomorrow’s brief.
One real decision from a recorded paper session on simulated prices, with the Critic’s verdicts written down but not applied. The company is not named and prices are left out. Step through it, or scroll.
What you control
Three decisions are yours. Everything else follows from them.
You set them once, change them whenever you like, and every change is written down with your name and the time.
- Posture
- How hard the desk pushes: how big a position may get, how much of the book is invested, the largest single order, the daily loss and drawdown halts. Five presets, the boldest of them on paper money only, or limit by limit.
- Autonomy
- Whether it asks you first, and where your line sits, the dollar amount above which nothing moves without your tap. On Manual every order waits for you; on Semi only orders over your line wait and the rest go on their own; on Full nothing waits, and Halt stops it at once.
- Guidance
- Plain-English notes to the Portfolio Manager: avoid a sector, favor a theme, sit out an earnings week. It must acknowledge each one in writing at its next session.
Every limit applies in every mode. The mode only decides who clicks.
Your line
You set the line.
Your line is the dollar amount above which no order goes out without your tap. You also set the limits, the risk appetite, how much the desk may do on its own, and any standing guidance. The desk works inside all of it and cannot loosen any of it. The drawing shows one day of proposals passing the same checks in the same order. Your line decides which ones wait for you.
Proposals drift in from the left, larger ones higher, pass the same 18 checks in the same order and come out three ways: waiting for you, going through, or stopped by a check. Where your line sits decides the split.
On Semi, what is above the line waits and the rest goes through. Here, about 4 in 10 wait.
Drag the line or switch the mode and the split changes. It is a drawing of the flow, not live orders.
IllustrativeDesk versus rules
Evidence, honestly.
The same signals and the same limits run twice, forward, every week: once by the desk and once by a rules twin, a copy of the book run by fixed rules only, kept for comparison. The twin already runs beside the desk; the gap between them is published every week, whichever way it points, starting with the first full trading week both have closed. Until then the chart below is a placeholder that only shows the shape of that report.
Example week 16: a counted week. The desk and the rules twin both traded, so the gap between them is published whichever way it points; in this drawing the desk finished ahead of the twin.
No real weeks yet. A gap as small as the one drawn here would need about thirty counted weeks before it meant anything either way.
Weekly gap, desk minus rules twin, in points. Tap or hover over a bar to read what kind of week it stands for.
What we test
Every strategy ships with an evidence page, and the page shows the misses. One published pivot method, replayed literally through the real risk limits over twenty years of daily prices, lagged simply holding the index; we published that and kept the method as one vote among several.
What it does not prove
A replay is not a forecast. It tests the rules and the limits, not the agents’ judgment, which is scored separately on trading days the model has never seen; costs, taxes and your own decisions will differ.
Built in, not bolted on
Safety by construction
The rules that matter are not instructions to a model. They are things the software has no way to do, limits enforced where the order is sent, and a switch that is always one click away.
Paper first
Every book starts on a paper account: real prices, pretend money. Going live is a deliberate, multi-step change made outside the console after weeks of paper evidence. Never a toggle.
Only the Executor holds keys
No agent has broker credentials, a network tool or a way to write an order. The one component that can trade has no model in it and is covered by tests line for line.
Halts and limits live in code
Daily loss, drawdown, exposure, order size, stale quotes, no trading around the open and close: checked at send time, every time. Loosening a limit asks for your authenticator code and leaves an audit line. Halt is one click from any screen.
Your book is yours alone
Each book is kept apart from every other book where the data lives: someone else’s agents, data and settings cannot read or touch yours, and a read-only guest sees numbers, never controls.
Nothing moves money
There is no code that can deposit, withdraw or link a bank. Funding happens at your broker, by you. Bellwether never sees bank details.
The checks
Every order, every time: the Executor’s checks in the order they run. The first failure wins and nothing is skipped.
- 01Paper account matches this desk
- 02Market open, outside the no-trade window
- 03Symbol passes the tradability screen
- 04Buy or sell only (no shorting)
- 05Limit order, good for today
- 06Limit price close to the current quote
- 07Quote is fresh
- 08Under the largest‑order limit
- 09Position stays under its size limit
- 10Book stays under its exposure limit
- 11Under today's order count limit
- 12Day-trading rule respected
- 13No earnings in the blackout window
- 14Daily loss limit not hit
- 15Drawdown limit not hit
- 16Not a duplicate of a recent order
- 17Over your line, it waits for you
- 18Settled cash covers it
FAQ
Questions
What is Bellwether, in one sentence?
A small desk of Claude agents that research, propose and review trades in US stocks and ETFs, one deterministic program that alone can place an order inside limits you set, and a console where you approve, give guidance and read everything they wrote.
Does it trade real money?
Not by default. Every book runs on a paper account with real prices. Going live requires weeks of paper evidence and a deliberate, multi-step change we make outside the console, never a switch inside it; if the book and the account it is connected to ever disagree about whether they are paper or live, the Executor halts rather than guess.
What do I actually control?
Three things, all from one screen. Your posture is a risk preset whose hard limits only move within fixed bounds. Your autonomy says whether the desk asks you first, and includes your line, the dollar amount above which nothing moves without your tap. Your guidance is plain-English notes the Portfolio Manager must acknowledge in writing. And from any screen, a Halt button stops new orders within seconds.
How do you test it?
Three ways, all published in the console. Rules and the risk policy are replayed walk-forward over twenty years of daily data with costs, and the evidence page lists what the replay does not cover. Agent sessions are run on trading days after the model’s training cut-off and scored against what happened, including how well stated confidence was calibrated. And every live paper day is journaled, reconciled to the cent and reviewed by the Critic. None of this predicts anything; it tells you honestly what happened.
How does it get better over time?
Every session is journaled and every proposal is later scored against what the market actually did. After the close the desk reviews itself: lessons that keep earning their place are kept, the rest are retired, and the Critic’s track record is checked for calibration: when it says 70%, is it right about 70% of the time? Changes to how the desk trades are never silent: each one is versioned, explained, and revertible by you. Inside the console, a Learning page shows the grades, the mistakes that repeat and the lessons kept or retired, so you can watch the desk change over time.
What does it trade, and through which broker?
Long-only US-listed stocks and ETFs with day limit orders in regular hours: no shorting, no options, no crypto, no borrowing on margin, and total exposure capped at your equity or below it. During the beta it works with one US broker that offers free paper trading; you bring your own account and connect it yourself after signing in. The trading key is stored sealed and Bellwether never asks for anything that can move money.
Who can see my book?
You, and anyone you explicitly invite as a read-only guest. Books are kept apart from one another where the data lives: no other book’s agents or people can read yours. Whoever runs Bellwether for you administers sign-ins and invitations, never your positions or your settings.
How do I get in?
Join the waitlist. We are letting people in slowly while the desk runs on paper; when a seat opens you get a one-time setup link by email, choose a password and enroll an authenticator app. It runs in the browser on a computer or a phone: nothing to download, nothing to install.
Our goal is the most intelligent trading desk in the world, one that works for you around the clock.
Today it runs on paper money in the open, and we are letting people in a few at a time. Leave your email and we will write when a seat opens.