Home / Guides / AI trading bot
AI trading bot: what the AI really does
Many tools call themselves an AI trading bot. In M Alpha Core the AI has a clear and limited job, and it is not predicting prices. This page explains what the built in assistant does, what it does not do and why the trading itself follows fixed rules.
Last updated:

The short answer
M Alpha Core has two parts that are often mixed up when people talk about AI trading:
- The trading follows one built in strategy, M Alpha Signal, with fixed rules. The scan finds the best settings of that built in strategy for your market and tests them on data the search never saw. Our signal server calculates the strategy with exactly those settings at every candle close, and your bot places the orders. No AI decides a trade.
- The assistant is Claude, an AI assistant from the company Anthropic, built into the app. It explains every screen, answers your questions and tells you the next step of the setup. It does not trade, it does not choose settings for you and it does not forecast the market.
We built it this way on purpose. The rest of this page explains why.
What "AI trading bot" usually means
The term covers very different things. Some tools use it for a bot with fixed rules and a chat window. Some use it for a model that has learned patterns from past prices. Some use it for a bot that asks a language model whether to buy or sell. The word AI alone tells you nothing about how a bot decides, how it was tested or how it handles risk.
So when you look at any AI trading bot, ask three plain questions. Which part decides the trades? Has that part been tested on data it never saw, with the results shown to you? And what is the worst move a single trade went through, and at which leverage would it have been liquidated? If a tool cannot answer these questions in numbers, the label does not matter.
Why M Alpha Core does not let AI predict prices
A language model is very good with words. It is not a reliable forecaster of prices, and it does not claim to be. Its answers can change from one question to the next, and you cannot run a model's opinion over a year of past candles and get a clean backtest of it. A trading decision that cannot be tested cannot be trusted with real money.
Fixed rules can be tested. The same rules on the same candles always give the same trades. That is why the scan can try many settings on the first 65 % of a year, run the best ones on the last 35 % and show you how they held up. It is also why the report can show the worst move against a single trade and the leverage limit that follows from it. All of that depends on rules that do not change their mind.
The scan does use automated search: it tries many combinations of settings and keeps the ones that did best. That is optimisation, and it is honest to call it that. It is not a machine that knows where the price will go. The methodology page describes the search and the check on unseen data in detail.
What the assistant does
- Guides the setup. From the first start to the first running bot: license, server, Telegram, exchange key, scan, report, bot. It names the next open step and where in the app you find it.
- Explains the report. What a profit factor of 2.5 means, why the unseen part matters, what the worst move against a trade says about leverage, what isolated and cross margin change.
- Explains errors. If a bot reports a problem, the assistant explains the usual cause and what to check, for example an expired API wallet on Hyperliquid or a key without trading rights.
- Answers in your language. You can ask in English, German, Spanish or any other language the model understands, by typing or by voice.
- Knows the app. A handbook of the app is built in and goes with every question, so the assistant names buttons and screens that really exist.

What the assistant does not do
- It does not trade. It cannot place, change or close an order. Your bot does that, following the strategy.
- It does not click for you. You press the buttons. The app does the work behind them.
- It does not predict prices. It will explain what a report says. It will not tell you where Bitcoin is going tomorrow.
- It does not decide for you. Leverage, margin mode, stake and market are your choices. The assistant explains what they mean.
- It does not want your keys. Exchange keys, passwords and tokens belong only in the protected fields of the app. If you paste one into the chat, the assistant tells you to replace it.
What the assistant knows about you
With every question, the app sends the built in handbook and a short status of your setup: for example whether a server is connected, whether Telegram is set up and how many bots and saved analyses you have. That status contains only yes or no answers and counts. It contains no names, no addresses, no amounts and no keys. Anything beyond that, the assistant only knows if you write it yourself.
The assistant runs with your own key at Anthropic, so your questions go from your device to Anthropic, not through our servers. You top up about $5 of credit once. With normal use it lasts around a year. The security page describes how the rest of the app handles keys and data.
How the trading itself works
- You choose a market and a time frame, the 30 minute or the 1 hour chart.
- The scan finds the best settings of the built in strategy for that market on the first 65 % of a year.
- The settings run once on the last 35 %, which the search never saw. The report shows both parts.
- You start a bot with exactly those settings on your own server. It first runs as a dry run.
- Our signal server calculates the strategy at every candle close and hands a sealed signal to your bot.
- Your bot places the order on your exchange with your trading only key.
Every step is rule based and repeatable. The assistant sits beside this process and helps you through it. It is never inside it.

Real numbers instead of AI promises
Because the trading follows tested rules, we can show real figures from our scans. Gold on the 1 hour chart: 153 trades in the test period, a profit factor of 2.53 on the searched part and 2.81 on the unseen part. An ETH example: 116 trades, a win rate of 58.6 %, a worst move of 15.64 % against a single trade and a leverage limit of 6.39x. These are backtests on past data, not promises, and the report shows the bad parts as plainly as the good ones. More are on the gold trading bot, Ethereum trading bot and S&P 500 trading bot pages.
Red flags with AI trading offers
- Guaranteed or fixed returns. No trading method can promise a return. A claim like that is a warning sign by itself.
- No test on unseen data. A curve that was fitted on the same data it is shown on proves little.
- No worst case. If the offer shows profits but not the worst move against a trade or the deepest drawdown, the risk is hidden.
- Your money on their platform. With M Alpha Core your money stays on your own exchange account, and the bot's key cannot withdraw.
- A black box. If nobody can tell you which rules a trade followed, you cannot judge whether to trust the next one.
The guide on how to test a crypto trading bot turns these points into a checklist.
Frequently asked questions
Is M Alpha Core an AI trading bot?
It has an AI assistant built in, but the trading follows a fixed, rule based strategy whose settings are tested on unseen data. No AI decides a trade.
Which AI does the assistant use?
Claude, an AI assistant from the company Anthropic. You use it with your own key and about $5 of credit, which lasts around a year with normal use.
Does the AI predict prices?
No. The assistant explains the app and the report. It does not forecast prices and does not decide trades.
Can the assistant place trades?
No. It cannot place, change or close orders. Your bot trades, following the strategy.
What does the assistant know about my account?
Only a short status with yes or no answers and counts, such as how many bots you have. No names, addresses, amounts or keys.
Do my questions go through your servers?
No. The assistant runs with your own key at Anthropic, so your questions go from your device to Anthropic.
Why not let AI decide the trades?
Fixed rules can be tested on a year of data and on a part the search never saw. An AI opinion cannot be tested that way.
Is the scan AI?
The scan is an automated search that tries many settings of the built in strategy and keeps the best. It is optimisation, not a prediction of prices.
Do I need the assistant?
It is the easiest way through the setup and the report, and it answers questions in your language.
Is AI trading profitable?
No label makes a strategy profitable. What matters is a test on unseen data, the worst case in numbers and your own risk control.
Compare M Alpha Core
Side by side with other bot tools, with sources for every statement about them.
Exchanges in M Alpha Core: Binance · Bybit · Bitget · OKX · MEXC · Gate · BingX · KuCoin · Crypto.com · Hyperliquid
Related pages
All figures come from backtests on past market data. No result is a promise of future profit. Trading, especially with leverage, involves high risk. M Alpha Core is a software tool, not investment advice.