QuantAgent introduces a novel multi-agent LLM framework tailored for high-frequency trading (HFT), addressing the limitations of existing LLM-based financial systems that rely on slow, text-driven inputs. The framework decomposes the trading process into four specialized agents—IndicatorAgent, PatternAgent, TrendAgent, and RiskAgent—each operating on structured OHLC price data. IndicatorAgent extracts compact technical signals (RSI, MACD, etc.), PatternAgent uses multimodal reasoning to detect chart patterns, TrendAgent identifies directional bias via slope-aware trend detection, and RiskAgent integrates all signals into a risk-reward framework with fixed stop-loss and LLM-predicted take-profit levels. The system is built on LangGraph and generates traceable, natural-language rationales for each trade decision.

Extensive experiments across nine financial instruments (including Bitcoin, Nasdaq futures, S&P 500, and VIX) at 1-hour and 4-hour intervals demonstrate that QuantAgent consistently outperforms random, linear regression, and XGBoost baselines in both directional accuracy and risk-adjusted returns. Notably, it achieves up to 80% directional accuracy in a case study on SPX and shows robust risk control, with the best or near-best maximum and minimum returns across most assets. The multi-agent structure enhances interpretability and robustness through cross-agent validation, making QuantAgent a scalable and transparent solution for real-time, data-efficient HFT systems. Limitations include reduced accuracy on ultra-short timeframes (1-15 minutes) and latency issues, suggesting future work on streamlining tool orchestration and edge deployment.