Implementation:HKUDS AI Trader A Stock Agent Config

Knowledge Sources	HKUDS_AI_Trader
Domains	Configuration, A_Shares, Agent_Setup
Last Updated	2026-02-09 14:00 GMT

Overview

JSON configuration files that define agent parameters, model selections, date ranges, and logging paths for Chinese A-share and generic daily/hourly trading experiments.

Description

Four user-facing configuration files provide the parameterization for running AI trading agents across different market types and time granularities. Each file specifies the agent class to use (BaseAgentAStock, BaseAgentAStock_Hour, BaseAgent, or BaseAgent_Hour), a date range for backtesting, a list of LLM models with enable/disable flags, agent hyperparameters (max steps, retries, initial cash), and a log output path. These complement the existing default configs (default_config.json, default_astock_config.json, default_crypto_config.json) which are already documented.

Usage

Pass these config file paths to main.py via the --config CLI argument. The load_config() function reads the JSON and uses its values to configure the agent run. Users edit these files to select which models to run, adjust date ranges, and set initial capital.

Code Reference

Source Location

• Repository: HKUDS_AI_Trader
• Directory: configs/
• Files:
  • configs/astock_config.json (62 lines)
  • configs/astock_hour_config.json (64 lines)
  • configs/default_day_config.json (51 lines)
  • configs/default_hour_config.json (51 lines)

Schema

{
  "agent_type": "BaseAgentAStock",
  "market": "cn",
  "date_range": {
    "init_date": "2025-10-01",
    "end_date": "2025-10-29"
  },
  "models": [
    {
      "name": "claude-3.7-sonnet",
      "basemodel": "claude-3-7-sonnet-20250219",
      "signature": "claude-3.7-sonnet",
      "enabled": false
    }
  ],
  "agent_config": {
    "max_steps": 30,
    "max_retries": 3,
    "base_delay": 1.0,
    "initial_cash": 100000.0
  },
  "log_config": {
    "log_path": "./data/agent_data_astock"
  }
}

I/O Contract

Inputs

Name	Type	Required	Description
(none)	—	—	Static config files; no runtime inputs

Outputs

Field	Type	Description
agent_type	string	Agent class name to instantiate
market	string	Market identifier ("cn" for A-shares; absent for US)
date_range	object	init_date and end_date for backtesting window
models	array	List of LLM model configs with enabled flags
agent_config	object	max_steps, max_retries, base_delay, initial_cash
log_config	object	log_path for output directory

Usage Examples

# Run A-share daily backtesting with astock_config.json
python main.py --config configs/astock_config.json

# Run US stock hourly backtesting
python main.py --config configs/default_hour_config.json

Related Pages

• Environment:HKUDS_AI_Trader_Python_LangChain_Runtime