I built a real-time trading control panel where multiple AI agents, powered by a local LLM, work together on trade decisions.

For a long time, my trading algorithms relied on predefined logic blocks and filters that I built myself. That approach can work, but markets are never static — they keep changing. And because of that, I found myself constantly stepping in to tweak the algorithm so it could stay aligned with current conditions. In the end, the whole system depended too much on me being involved all the time.

https://preview.redd.it/tgf8xctzz2ng1.png?width=1499&format=png&auto=webp&s=fe81ab005dd929799ac0273f2196b4f5346afee4

So I wanted to build something different — a system that could analyze the market, challenge ideas, deal with uncertainty, manage risk, and monitor positions more like a small team of traders than a single hard-coded script.

That is how this project came together.

At the center of it is a control panel that shows the state of active bots and open positions in real time. But it is more than just a dashboard — it is the place where the full decision-making process comes together.

Behind it, several agents work as a team:

• one analyzes market structure and looks for a valid setup;
• another plays the role of a skeptic and filters out weaker signals;
• a third focuses on risk, position sizing, and protective logic;
• and a fourth monitors the open trade and decides whether the position should be held or closed.

So instead of relying on one simple “buy/sell” trigger, each trade goes through a multi-layer decision-making process.

What I like most is that this system is not just built to find entries. It is also designed to avoid weak trades, handle uncertainty, and make the reasoning behind decisions visible through the control panel. And most importantly, I no longer have to constantly reshape the algorithm by hand to keep up with the market — part of that adaptation is now handled by the agents themselves.

To me, this feels much closer to the future of algorithmic trading than traditional one-layer bots.

I’d love to hear thoughts from traders, quant developers, and anyone working with agent-based AI systems.