I still remember clearly that night, September 5, 2026, after testing the AI analysis system on the terminal for several weeks, I received an email from one of the customers using the iCafeFX app writing, "Your system isn't sending out any trading commands at all, just analyzing for me to listen. Here's the response from the system" along with a text file he had recorded from my system. I was greatly relieved that everything was working as intended. This confirmed that our safety-first approach had succeeded in creating a system that provided valuable insights without risking unintended trades.
Why Build an AI Analysis System on the Terminal
To begin with, I had previously tried using FLUX and Typhoon models on a Mac Mini M4 Pro, but found that chat-based question answering wasn't suitable for terminal trading that requires quick and precise responses. The transition from a conversational interface to a terminal-based system was necessary to meet the specific demands of trading environments. I needed a system that could directly connect to MT5 and analyze charts, while also ensuring it couldn't send commands to the trading machine itself. Building the system on the terminal provided more control than using a front-end application, allowing us to implement strict safety protocols while maintaining the responsiveness required for trading analysis.
Designing for Safety
What I emphasized most was safety. I didn't want the AI to have the authority to send trading commands outright. Therefore, I designed the system with two main constraints: first, the system would only respond through file channels and HTTP read-only methods, with no input reception that could lead to command sending. This architectural decision created a clear separation between analysis and execution, which was crucial for our safety objectives. Second, we use Redhat MCP AI as an intermediary for processing, which has mechanisms to prevent the direct writing of trading code. This additional layer of security ensured that even if the system were compromised, it couldn't execute trades without human intervention.
Answering Thai Questions with MT5 Charts
Making the system understand both Thai language and MT5 charts simultaneously was the most challenging aspect. I had to create a system that could analyze data from charts and convert it into easily understandable Thai text, particularly explaining trends, signal strength levels, and useful advice for traders. The process involved developing specialized algorithms that could interpret complex market data and translate it into natural language that Thai traders could comprehend intuitively. I didn't want it to provide suggestions that were bonuses or guaranteed profits, but rather focused on analysis and providing information. This approach maintained ethical standards while delivering valuable insights to users.
System Testing Process
Testing this system took over two weeks. We tested it by having the system answer the most complex questions, both technical and economic matters related to gold trading. I set a goal for the system to respond within 3 seconds, with answers approximately 1,748 bytes in length, which is an appropriate size for comprehensive information that isn't too lengthy. The testing phase involved multiple iterations to refine the system's accuracy and response time, ensuring it could handle various market conditions and query types without compromising on performance or safety.
Results from the System
After testing was complete, I received very satisfactory responses. The system could accurately analyze MT5 charts, translate them into clear Thai, and provide useful information to traders. What gave me confidence was that we verified the system's source code and found no order-sending commands whatsoever, which was our primary objective from the beginning. The successful implementation demonstrated that it was possible to create an AI system that could enhance trading capabilities without introducing unacceptable risks.
Lessons Learned and Next Steps
This project was an important creative endeavor for our team. We learned that when designing AI systems for trading, safety must be the top priority, along with creating systems that can work seamlessly with existing software. The experience showed us that careful architectural planning and rigorous testing could prevent potential security issues while delivering valuable functionality. In the future, we want to develop the system to enable better real-time analysis and enhance its ability to make more accurate trend predictions, building upon the solid foundation we've established with this initial implementation.
For those who want to try real gold trading, open an XM account at: https://clicks.pipaffiliates.com/c?c=72816&l=th&p=6
