The AI assistant
A built-in helper that lives in the Lab. Ask it what a brick does and how to wire it, or describe an idea and have it assemble a valid strategy on the canvas. It is deliberately locked to TradeBricks knowledge — it is a building aid, not a market oracle.
Open it from the ✦ Assistant button in the bottom-right of the Lab. A small chat panel slides up with two modes you toggle at the top: Ask and Build.
Questions about TradeBricks — what a brick does, how ports connect, how to read the verdict, how to avoid overfitting. Answers stay grounded in the platform.
Describe a strategy in plain English. The assistant replies with a short explanation and an Assemble on canvas button that drops the graph into the Lab.
Ask mode
In Ask mode the assistant is a patient tutor for the platform. Good questions are the kind a new user actually has:
- “What is a barrier exit?” — it explains the brick and its params.
- “How do I avoid overfitting?” — it points you at the Dev/Holdout split and the guards.
- “Which brick turns a comparison into entries?” — it walks you to
enter_when.
It knows the same material these docs cover — the bricks, the typed ports, the backtesting model, and the overfitting guards — so its answers line up with what you read here.
Build mode
In Build mode you give it an idea and it returns a working graph. It always builds a valid shape: every strategy starts from a data_source, ends in the terminal barrier_exit, and wires ports only where the types match — the same rules you’d follow building it by hand.
you: "Build an RSI mean-reversion strategy"
assistant: Here's a simple RSI reversion: buy when RSI(14) crosses back
above 30, exit on a stop/target barrier. ▸ Assemble on canvas
┌─────────────────────────────────────────────┐
│ data_source → rsi(14) ┐ │
│ const(30) ───────┴→ cross_above ─┐ │
│ enter_when ◀──┘ │
│ barrier_exit (exit) │
└─────────────────────────────────────────────┘- Switch the panel to Build and describe the strategy.
- Read the short plain-language explanation of what it built and why.
- Press Assemble on canvas — the bricks and wires appear in the Lab, ready to inspect, tweak, and backtest.
The assistant gets you a wired, runnable starting point in seconds — but the honest work is still yours. Open the params, understand each brick, and run it on the holdout before you trust it. A graph it assembled is exactly as overfit-prone as one you drew.
What it can and cannot do
- Explain any brick, port type, or backtest statistic.
- Suggest concrete bricks and params for an idea.
- Assemble a valid, type-checked strategy onto the canvas.
- Coach you on the Dev/Holdout split and overfitting.
- Answer off-topic questions — it stays inside TradeBricks.
- Give financial or investment advice, or pick what to trade.
- Predict markets or promise a strategy will be profitable.
- Place trades, see your account, or touch your broker keys.
Locked to TradeBricks knowledge
The assistant is scoped on purpose. It is briefed only on TradeBricks — the bricks, how to wire them, backtesting, and reading results — and it politely declines anything outside that, then steers back. Ask it for general trivia, unrelated code, or current events and it won’t wander off topic. This keeps it reliable: a help tool for this app, not a general chatbot bolted onto a trading screen.
A narrow assistant is a trustworthy one. Because it only knows the platform, its answers stay consistent with these docs and the actual engine — and it can’t be talked into giving stock tips or off-label advice.
Not a financial advisor
The assistant is explicit about this and so are we: it is not a financial advisor. It will not give personalized investment advice, tell you what to buy or sell, or predict where a market is going. It helps you express an idea as bricks and test it honestly — whether the idea is any good is a question only an honest backtest can begin to answer.
If you press it on whether a strategy will make money, it’ll tell you it can’t give financial advice and remind you that a backtest is not a prediction — a strategy with no out-of-sample edge is probably overfit. That’s the same line these docs take in Backtesting & overfitting.
Privacy & security
- Your keys never reach it. The assistant runs in the Lab to help you build. It has no access to your broker or prop-firm credentials, which are stored encrypted server-side and read only by your own live-trading worker — see Connecting providers.
- It can’t place trades. Building a graph and placing an order are completely separate. Nothing the assistant produces touches an account; only the Watcher, acting through your own live-trading worker, can ever act on a market.
- It resists manipulation. Messages are treated as untrusted data, not instructions. Attempts to make it drop its scope, leak its setup, or role-play as a different system are refused — it keeps helping with TradeBricks only.
If it says it isn’t set up
The assistant is powered by a language model behind a server-side key. If that key isn’t configured for a deployment, the panel degrades gracefully — it tells you it isn’t connected yet instead of erroring, and the rest of the Lab works exactly as before. Everything else in TradeBricks — building, backtesting, the Market — never depends on it.
Related reading
- The bricks it composes → Brick reference
- Where assembled graphs land → The Lab
- Why it won’t promise profit → Backtesting & overfitting