Ask your desk
anything.
Flux AI reads the tape, the filings and your own portfolio — then answers in plain English, with the numbers and the backtest attached. Six models keep working while you're away.
A working demo running on simulated desk data. In the app it reads your real positions and live prices.
Not a chatbot bolted onto a chart.
Flux AI is wired into the same engines, order book and portfolio the platform runs on. It answers from your actual state, not a generic model of the market.
Reads your portfolio
"What's my biggest risk?" returns your worst position by name, its size, live P&L and a Monte-Carlo VaR number.
Runs the models
Ask for a forecast and it calls PriceNet, checks RSI and regime, and hands back a number with a confidence score.
Backs it with history
Every setup can be replayed across years of sessions before you act. Win rate, Sharpe, and whether the edge is decaying.
Writes your reports
Morning brief, midday pulse, portfolio review — composed from live state on a schedule, or on demand.
Refuses to guess
Ask "should I buy?" and it stacks the evidence instead of pretending to know. It never gives advice.
Ends in an action
Every answer carries buttons — load the chart, paper trade, flag it, run the backtest. No dead ends.
Six models that never clock out.
They run on their own cadence all day — forecasting, classifying the regime, scoring headlines, simulating your book. You can pause any of them.
Live simulation of the agent loop. Model names describe the technique each agent uses.
Reports that write themselves.
The Reporter agent publishes on a schedule — built from your live positions, the day's movers, model forecasts and open risk. You wake up to the work already done.
Overnight moves, the day's catalysts, and what changed in your book while you slept.
What the engines surfaced since the open, re-ranked by confidence.
Where you finished, realized and open P&L, and the setups carrying into tomorrow.
Illustrative notes · synced from your Obsidian vault
Give it a memory. Yours.
Connect your Obsidian vault and Flux reads your theses, rules and trade journal as long-term memory. Its answers start sounding like you — it remembers your playbook, flags when you're about to repeat a mistake, and cites your own notes back to you.
- ❖Bring your notes. Your Obsidian vault, journal and watchlists become the AI's context.
- ◇It remembers your rules. "You said max 8% per name — this order is 11%."
- ⊘Private by design. Your vault stays yours; live sync runs through the desktop app.
Ask Flux to research any company.
Type a ticker and Flux assembles the deep-dive — the business, the bull and bear case, the numbers and the catalysts — in one page. Every figure below is a simulated sample, not live data.
How an answer gets built.
Read state
Your positions, orders, watchlist and the live tape — the question is grounded in what's actually true right now.
Call the models
Forecast, regime, sentiment and risk models run against the symbols in scope.
Check history
Any claimed edge is replayed across years of sessions before it's put in front of you.
Answer + act
Plain English, the supporting numbers, a confidence score, and the buttons to do something about it.
It works for you.
Not the other way around.
An AI with access to a trading account is only useful if it's bounded. Flux AI is built to inform and execute what you approve — never to act on its own.
No order is placed without you confirming it. There is no autonomous execution mode.
Position size, stop distance, correlation and exposure are checked before anything routes.
Flux connects with your credentials. We never take custody of your funds.
Every question, model run, report and order is written to an activity trail you can audit.
The catch: IV at 62% means the move is largely priced. Backtest on this setup is 71% over 5 years — but that's history, not a promise.
Flux is research software, not a financial adviser.
What people ask about the AI.
Does Flux AI place trades by itself?
No. It researches, forecasts and drafts orders, but nothing routes until you confirm it. There is no setting that lets it trade autonomously — that's a deliberate design decision, not a missing feature.
Is this just ChatGPT with a stock prompt?
No. The language layer is one part of it. Underneath are six purpose-built models — a sequence forecaster, a volatility-regime classifier, a sentiment scorer, a Monte-Carlo portfolio simulator, a signal re-ranker, and a report writer — plus direct access to your live positions and the engines' output.
Can it predict the market?
No, and it won't claim to. It produces probabilistic forecasts with explicit confidence scores, and shows you the historical base rate behind any setup. Confidence is a measure of model agreement, not a guarantee.
What does it actually read?
Price and volume history, the options tape, SEC filings and 8-Ks, earnings material, the news wire, and your own portfolio and order history. It cites which of those a given answer drew on.
Where does my data go?
Your positions and activity stay tied to your account. Flux connects to your broker with your own credentials and never takes custody of funds.
Put six models on your desk tonight.
They'll have been working for hours by the time you open it.