WTS AI Coach

What this is

A trading study program for Windows with two halves: a chat that answers questions out of the handbooks, and a learning platform that walks through a book and keeps track of where the learner stands.

What sets it apart from a general chatbot is the source. Ask a language model about order flow and you get the average of what the internet says about trading, contradictions included. The coach draws its answers from one library: the five volumes whose method the Trading Station itself is built on.

What it costs

The coach is not free. To try it out there is a test account with five free questions; after that the program stops accepting input until the account is unlocked. Unlocking happens on request — a message is enough.

Two things worth knowing beforehand:

• The five questions apply to the whole program, not per window. The chat and the learning platform draw on the same allowance, so an exercise in the learning path uses one up.
• A test account sends a shorter conversation history than an unlocked one. Anyone who wants a long chain of follow-up questions will notice.
• What is counted is a call to the model, not an exam question. The placement test asks its ten questions in one go and therefore costs two: one for the questions, one for marking the answers.

What five questions are good for: installing the program, creating an account, looking at the chat and the learning path once, and checking whether answers drawn from the handbooks suit you. They are not enough to work through a volume, and that is deliberate.

The account

The program does not run without signing in. You create the account inside the program, not on this page: below the ANMELDEN button there is a text link that switches the form to TESTVERSION ANLEGEN. Three fields appear.

One thing to know before you install: the program's own interface is German only. The buttons, the messages and the study path are not translated. The quoted labels and errors below are therefore the German ones, with the meaning next to them.

Little is asked for, and that is deliberate:

User name	two to thirty-two characters — letters, digits, dot, underscore, hyphen, at least one of them a letter or digit
Password	at least eight characters, entered twice
Email address	not asked for

The account is free and unlocks the five free questions. It is the same account you sign in with after unlocking — it gets extended, not replaced.

One account per computer

Signing up sends along an identifier for this machine, and the server binds exactly one test account to it. Creating a second one on the same machine is refused — as long as the first is still a test account. Unlocking releases the binding, because the brake against free second accounts is then pointless.

Two things about that, so it does not sound like more than it is. The identifier is hashed before it is sent; what leaves the machine is a checksum, not a hardware ID. And a brake is a brake: it comes from the program and can be forged. It stops the convenient second test account, not the determined one.

When it does not work

• „Auf diesem Rechner gibt es bereits ein Testkonto." — one was already created on this device. Sign in with it, or write to us.
• „Die Registrierung ist zurzeit geschlossen." — self sign-up is switched off at the moment. We then set the account up by hand.
• „Zu viele Fehlversuche — bitte in 10 Minuten erneut versuchen." — there is a lock against automated sign-ups, per connection and in total. If a whole classroom sits behind one connection, the fourth person waits ten minutes.

How it knows my trading style

Not by watching, but by reading. The five handbooks exist as PDFs and are read in once before the first question: pull out the text, cut it into overlapping pieces, turn every piece into a vector. Some fifteen hundred pages become a few thousand searchable passages.

For every question the program finds the five closest passages and puts them in front of the question, with the instruction to stay with them and follow that method. The window header states how many passages are in the store; below the answer it says whether book knowledge went into it.

That is the whole trick, and it has an advantage you only notice on re-reading: the coach cannot teach the opposite of what the books say — it has nothing else to hand.

Volume	Focus
Trading Basics	contracts, tick and point value, order types, risk per position
Advanced Order Flow	footprint, delta, absorption, imbalance, auction, order book
Models & Trading Strategies	order flow setups, research, risk
Trading Strategies and Analytics	building, testing and tuning your own strategies
AI Trading & Analytics	language models, Python, quant research

The volumes are described one by one under books.

How it knows the learner

The second half is the more interesting one. A textbook does not know who is sitting in front of it; the coach writes it down.

Placement. Before the first chapter of a new book come ten questions of rising difficulty — from the definition through a worked example to carrying an idea over into a scenario. The result does not decide about access, it decides the level things start at.

Study path. The book is broken into chapters, taken from the bookmarks of the PDF. Each chapter opens with a summary, then a question aimed at the weaknesses on record. The answer is marked: right, partly right or wrong, plus what is missing and a hint on request.

Learning profile. Every mark is written into a file per user: strengths, weaknesses, topics covered, history. That file feeds back into the generation of the next exercise — so the exercise can see where things went wrong last time, and it aims there.

Progress. Which book, which chapter, what is already done and how the placement went sits in a second file. Close the program and open it three days later, and it carries on where it stopped.

Both are text files — but they sit on the server, one set per account, not next to the program on your machine. The desktop program keeps none of it. Anyone who wants to see their learning profile, or have it deleted, asks the provider; it belongs to the account and goes with it.

Exercises that ask for something

The exercises are mixed and generated from the book knowledge, not taken from a fixed list:

• Comprehension — explain why absorption is not the same thing as an imbalance
• Practice — describe a chart or numbers scenario and place it
• Python — write a small calculation yourself, where the volume supports it

Marking comes with a score and a reason. The score is not a report card; it steers the next exercise.

What runs today and what does not

Today: chat with book knowledge, study path with chapters and summaries, placement test, exercises with marking, learning profile and progress per account.

Not yet: handing charts to the chat as images and having them discussed. Exercises taken from real recorded sessions instead of described scenarios — the recordings sit in WTS Analytics, the bridge to them is not built. Reading in your own trading journal and having it checked against the method. Saved conversations. Sources with book and page under every answer.

This list stands here because an announcement without a line between runs and planned is only a claim. What sits under "today" works — as a test version, with everything that implies.

What it is made of

A Windows program written in Python, shipped as an executable with an installer — but that is only one half. The other is a server: it holds the accounts, the ingested book knowledge, the learning profiles and the key for the language model. The program on your desk asks questions and shows answers; the computing does not happen there.

There is no model menu. Which language model answers is fixed and set on the server, not chosen by the user. The window header names the model for information, not for selection.

Where your questions go. Every question leaves your machine. It travels, together with the book passages the program puts in front of it, to the provider's server and from there to the operator of the language model (ollama.com). A model computing on your own machine does not exist in this version. Anyone who does not want that cannot use the coach in any meaningful way — that is put more bluntly than is customary, but it is the truth about this program.

What the server keeps about you

Because the privacy policy points here for this question, it is stated here rather than in the small print:

What	Where	How long
User name and password (as a verifier, never in clear text)	account on the server	until the account is deleted
Sign-in token for "stay signed in"	account on the server	90 days, otherwise 24 hours
Checksum of this machine	account on the server	as long as the test account exists
Learning profile and progress	files on the server, one set per account	until the account is deleted
The conversation itself	is not stored	—

Deletion and access requests go through the provider; the address is in the imprint.

What it is not for

The coach explains, questions and marks. It gives no investment advice, no signals and no setups to trade, and it does not say what the market will do next. A language model can also be wrong — even with a good book behind it. What it says is study material, not an assurance.

Trading futures carries the risk of total loss.