Showcase · Auto-Trader

A trading agent that reads the news first

Most trading bots react to a chart. Auto-Trader reads the cause: news and market data become signals, a brain weighs them, and a risk layer decides whether a position is allowed at all. Fully autonomous, paper-only, and every trade traceable back to the source that triggered it.

paper
only — no real money
24/7
news & data harvester
100%
trades traceable to source
0
positions outside the risk guardrails
From news item to position

Six steps, one log

Every position goes through the same chain. The brain may propose; the risk layer has the final word. No step happens without being traceable afterwards.

1

News & market data

A harvester continuously collects prices, news and sentiment — the raw input, with source.

2

Signal extraction

Noise stripped out: only events that matter for a position, with a weight and a source reference.

3

Reasoning brain

Model-agnostic. Weighs signal against existing positions and context, and proposes a trade with reasoning.

4

Risk guardrails

Position limits, exposure caps and stop-loss review every trade. Outside the limits? Rejected and logged.

5

Paper execution

The approved trade is executed in a paper portfolio — realistic, but without real money.

6

Log & learner

Every trade, piece of reasoning and outcome is logged and evaluated. The next decision is better informed.

Technology in detail

Discipline over conviction

Source-first, like everything else

No trade without a traceable cause. Every signal links back to the news item or data point that produced it — the same citation discipline as the rechtspraak RAG.

Model-agnostic brain

The reasoning layer is decoupled from the model: plug in an API, an EU-hosted model, or one running fully on-prem. The brain proposes — the risk layer decides whether it's allowed.

Risk is a guardrail, not advice

Position size, exposure per market and stop-loss are hard-enforced in code. A convincing prediction still doesn't open a position outside the limits.

Paper-only by design

The agent trades in a paper portfolio. This proves the strategy on real data without financial risk — and without Neuralex giving investment advice.

Learning loop

An evaluation after every session: what worked, what didn't, and why. That feedback sharpens the next decisions.

Fully traceable

Signal, reasoning, (rejected) trade and outcome sit as a single thread in the log. Every move can be explained after the fact.

Built with News/data harvesterSentiment extractionModel-agnostic LLM layerRisk enginePaper brokerEvaluation log
In development · paper-only

Same blueprint, on market data

Auto-Trader proves the Neuralex approach on one of the hardest domains: source-first, autonomous where possible, guardrails where necessary. Paper-only — no investment advice. Curious how this would work on your data?