QuantHarness provides a multi-agent LLМ system for price-driven high-frequency trading, with a Flask web interface and programmatic usage. It combines indicator, pattern, trend, and decision agents and supports multiple LLM providers via configuration.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Official Repository for QuantHarness. A sophisticated multi-agent trading analysis system that combines technical indicators, pattern recognition, and trend analysis using LangChain and LangGraph. The system provides both a web interface and programmatic access for comprehensive market analysis.
How it works
The project defines four agents:
- Indicator Agent: computes five technical indicators on each incoming K-line (e.g., RSI, MACD, Stochastic Oscillator).
- Pattern Agent: analyzes recent price charts to identify pattern matches and returns plain-language descriptions.
- Trend Agent: uses annotated K-line charts with fitted trend channels to quantify market direction, channel slope, and consolidation zones.
- Decision Agent: synthesizes outputs from all agents to formulate LONG or SHORT trade directives with entry/exit points and stop-loss thresholds. The Web Interface provides real-time market data from Yahoo Finance, asset selection, multiple timeframes, dynamic chart generation, and API key management.
Getting started
Installation steps from the README: "conda create -n quantharness python=3.11" "conda activate quantharness" "pip install -r requirements.txt" If TA-lib issues arise, "conda install -c conda-forge ta-lib".
Set up LLM API keys via environment variables: "export OPENAI_API_KEY="your_openai_api_key_here"" "export ANTHROPIC_API_KEY="your_anthropic_api_key_here"" "export DASHSCOPE_API_KEY="your_dashscope_api_key_here"" "export MINIMAX_API_KEY="your_minimax_api_key_here""
Usage to start the web interface: "python web_interface.py"
The web app is available at http://127.0.0.1:5000.
Python usage example: "from trading_graph import TradingGraph" "trading_graph = TradingGraph()" "initial_state = {\n "kline_data": your_dataframe_dict,\n "analysis_results": None,\n "messages": [],\n "time_frame": "4hour",\n "stock_name": "BTC"\n}" "final_state = trading_graph.graph.invoke(initial_state)"
Configuration options include:
- agent_llm_model (default: "gpt-4o-mini")
- graph_llm_model (default: "gpt-4o")
- agent_llm_temperature (default: 0.1)
- graph_llm_temperature (default: 0.1)
Full list of configurations is in default_config.py.
Recent releases
RELEASES (latest 0): - none
Traction
Stars: 2820 Forks: 604 Open issues: 5
License
MIT
Caveats
Licensed under MIT. Data sources include Yahoo Finance. The README notes support for multiple LLM providers (OpenAI, Anthropic, Qwen, MiniMax) and specific environment variable configuration for API keys. It mentions TA-Lib installation considerations and image-input requirement for the LLMs in this setup. The repository was created 2025-07-10 and last pushed 2026-07-24.