01CONTINU
FREUD can walk away from its position
At every decision the position is re-examined. If the reasons for holding it disappear, the agent can cut it before the stop, without waiting for the loss.
Freud dives into price interactions across four time scales, draws its decisions from them and reports on each one, before its outcome.
No payment · your funds stay with your broker · never any access to your withdrawals
several scales · one decision every 30 min
Sample line · decision-log format
Simulator cost model · paid once per trade
Illustrative trace · the pale traces are previous cycles
Over eight years, the best computable policy beats the costs of a raw account. It sets the target the agent must approach.
Gain of the best policy on a period never seen before, read only once. Positive, still too short to be confirmed.
At these costs the best computable policy does not trade. We do not offer it.
Published when validation ends.
Training under way
Each decision writes one line: what the agent saw, the action it preferred, the runner-up, and the gap between them. This terminal shows that format on a replayed session; its scenarios and probabilities are illustrative.
The log format, on a replayed session. The decisions in this demo are written by hand; the agent’s own will be published, timestamped, during the beta.
Join the beta →Before spending an hour of GPU, we compute the best possible policy for the information the agent receives and the costs it pays. If that ceiling does not beat the costs, no agent will — and we offer nothing on that market.
Measured ceilings · protocols written before the computation
The lower the cost, the stronger EUR/USD’s gain: a real signal, held back by cost. On gold, lowering the cost changes nothing.
Data 2016 → February 2024. Optimal policy computed on out-of-sample forecasts, walk-forward (T102, 28 Sept 2026). Newey-West statistic on daily gains.
| Cost (pips) | strength |
|---|---|
| 0.50 | +3.99 |
| 0.60 | +3.94 |
| 0.70 | +2.70 |
| 0.80 | +1.94 |
| 0.84 | +1.80 |
| 1.00 | +1.79 |
| 1.20 | +0.49 |
| 1.40 | +0.52 |
| 1.63 | −1.13 |
| 1.80 | −0.58 |
per decision: what the agent saw, what it preferred, and by what margin.
staying flat is a decision like any other.
2024 → 2026 judges the model once, without being used to choose it.
the EUR/USD agent on a demo account, with a public, timestamped log.
RISK MANAGEMENT THE LIFE OF A POSITION
The safeguards planned for the beta, in the order they kick in. The scenario below is an illustration.
At every decision the position is re-examined. If the reasons for holding it disappear, the agent can cut it before the stop, without waiting for the loss.
When the loss passes half the risk budget, severity rises — CAUTION then DANGER — and the size is cut.
The guard closes everything at 80% of the cap, before the limit, and freezes the session until conditions become readable again.
Automatic: a global cut that cannot be bypassed. And manual: you can stop everything, at any moment.
THE READING WHAT FREUD SEES
The current price says little about what comes next. Over ten years of EUR/USD, the usable information at thirty minutes is thin — and often smaller than the cost of a round trip.
FREUD reads the market like a retina: coarse on the distant past, fine near the present. Several time scales, one reading.
A perception module sums up what can be read, and measures the confusion of what cannot. The agent acts only if the expected gain exceeds the cost — otherwise, it does nothing.
From the dot-com crash to the inflation shock, the same policy runs through all nine periods. The last one, 2024 → 2026, is held back: it is used to choose nothing and judges the model only once.
Acceptance thresholds, written into the code
EUR/USD 2004 → 2026 · 215,699 thirty-minute blocks, built from one-minute bars.
Dot-com, 2008, flash crash, Brexit, Covid, inflation — and 2024 → 2026 held back.
No parameter adjusted along the way.
Run inside the image before every training: one failure, and nothing starts.
ROADMAP WHAT COMES NEXT
FREUD is the agent specialised on one instrument. AXON will extend the reading to four asset classes: the specialised agents will share their signals, and a meta-allocator will distribute capital while arbitrating the risk of the whole.
The agents spot opportunities and share their signals.
Each agent proposes its positions, capital is allocated proportionally.
The meta-allocator adjusts exposure to the swarm’s aggregate risk.
AXON v0 · 4 agents — planned 2027 · AXON v1 · 12 multi-asset agents — 2028 · not available today
FAQ COMMON QUESTIONS
An MT4/MT5-compatible broker account and an API key in “Read & Trade Only” mode. Basic trading knowledge is recommended.
Yes, absolutely. Your funds stay in your broker account. FREUD has no control over transfers — the separation is structural.
Nothing during the closed beta. No price is set until the agent has passed its validation.
Publishing the failures makes the rest credible. On gold, at retail costs, the best possible policy does not beat the costs: we measured it, and we do not offer it.
FREUD is software: it manages no money on behalf of others, and your capital stays with your broker. Before any order is placed on a client account, compliance (AMF, MiFID II) will be established with our legal advisers.
Losses are possible. During the beta, the agent runs on demo accounts; the safeguards described above limit losses without removing them. See the risk warning.
CLOSED BETA WAITING LIST
No payment · no account to create · your funds stay with you