Quick summary

  • Built an AI system that reads MT5 charts and answers questions in Thai, running directly on the terminal.
  • Designed with strict safety constraints: no order execution capability, read-only endpoints only.
  • Tested for 2+ weeks with complex technical and economic questions; responded within 3 seconds.

Why run AI analysis directly on the MT5 terminal?

I had previously tried running FLUX and Typhoon models on my Mac Mini M4 Pro, but found that chat-style interfaces weren't suitable for terminal trading. Traders need fast, precise answers—delivered where they're already working. Running the system directly on the terminal gave me far more control than using a separate client application.

The system connects to MT5 in real time, analyzes charts as they update, and provides immediate insights. This eliminates the friction of copying chart data to another tool or waiting for batch processing.

How do you keep an AI trading system safe?

Safety was the top priority from day one. The system has two hard constraints: it only responds through file output and read-only HTTP endpoints, with no input channel that could trigger order execution. We use Redhat MCP AI as the processing middleware, which has built-in safeguards against writing trading code directly.

After testing, I reviewed the source code and confirmed there were absolutely no order-sending commands anywhere. This was our primary goal from the start.

For a deeper look at EA safety, read 7 rules before letting an EA near your real portfolio.

Can an AI system understand both Thai and MT5 charts?

Yes, but it requires careful design. The system analyzes chart data and translates findings into clear Thai text—explaining trends, signal strength, and actionable insights without making profit guarantees or bonus promises.

Testing took over two weeks with the most complex questions possible, covering technical analysis and gold market economic factors. The target was responses within 3 seconds, around 1,748 bytes—comprehensive but not unwieldy.

The system performed well. A user confirmed it only analyzed and reported, never sent trade orders, which validated the safety design.

For more on our AI testing methodology, see AI vs. human analysts: 65-case backtest results.

What's next for terminal-based AI analysis?

We want to improve real-time analysis capabilities and enhance predictive accuracy for trend forecasting. The foundation is solid: a safe, responsive system that traders can trust.

If you're building or evaluating AI trading tools, start with safety constraints before adding features. A system that can't accidentally trade is worth more than one that can do everything but might.

For our gold analysis framework, check out the six factors that drive gold prices.