幻银量化A股AI交易系统A-share-Ai

基于四大国内大模型（幻银超i、DeepSeek、通义千问、豆包）的A股自动交易系统，支持全市场选股和实盘交易。

接入自产龙虾iClaw，手机端一句话同时控制四大模型全自动交易。

交易执行说明

• 买入：按总资产×仓位比例计算金额，并受「单轮最多使用可用现金一定比例」限制。
• 卖出（模型自动 SELL 信号）：按当前持仓全部卖出，不受可用现金上限影响（避免资金紧张时卖单被算成 0 股）。

功能特性

🎯 让AI在A股市场中一展身手，让多个大语言模型在A股全市场股票中完全自主决策、同台竞技！

🎯 接入A股真实交易账户，实时展示ai真实交易过程，实时+真实交易让系统有更广阔的应用空间

🎯 多模型集成：支持幻银超i、DeepSeek、通义千问、豆包等多种大语言模型

🎯 策略进化：基于交易表现，AI模型能够不断优化自己的投资策略

⏰ 秒级别实盘交易支持 - 接入实盘秒级精度交易,需拥有合规通道

🚀 服务部署与并行执行 - 部署生产服务 + 并行模型执行

🎨 ai交易逻辑展示 - 详细的交易日志可视化（完整交易过程展示）

🎯 核心特色: 100% AI自主决策，零人工干预，ai策略自我发现、自我训练、自我进化

🎯 特别功能：交易中可添加实时人工干预和Ai进行互动（可选），接入自产龙虾iClaw，接管互动过程并随时指令四大ai同时干活。

🎯 新增了异步执行功能，提高系统运行速度

🤖 4大模型并行：幻银超i、DeepSeek、Qwen通义千问、豆包同时运行，比较收益
📈 全市场选股：AI自主从全市场选择有潜力的股票，自定义股票池（可选）
💹 实时行情：集成实时A股数据，供ai使用并用于实际交易
💹 历史行情+财务数据：集成A股历史行情和财务数据，供ai分析和训练
🎯 智能交易：基于AI分析生成买卖信号并全自动执行交易
🌈 多种策略：包括趋势跟踪、均线策略、技术指标策略、量化套利策略等，完全由ai自动选择并不断进化
📊 实时监控：UI界面实时显示收益曲线和交易日志
🔒 风险控制：内置仓位控制和止损机制

📱 iClaw钉钉配置说明

方法1（已取消）：启用 config/dingtalk.json 后，程序启动会提示先在钉钉私聊机器人发送一次 /status 完成绑定；绑定成功会自动把你的账号ID写入 default_conversation_id，后续每轮交易完成将自动推送到你的钉钉私聊（无需手动配置个人ID）。

方法2：通过测试脚本或钉钉平台获取 sender_id 写入 config/dingtalk.json 的 default_conversation_id 字段，同时配置其他必要参数（见exampl文件），程序启动后会直接将每轮交易结果推送到该账号，同时实现钉钉对项目所有大模型的远端实时控制。

Stream 长连接：入站消息走 WebSocket Stream，出站通知走 REST API（二者独立）。若长时间收不到私聊指令但交易通知仍正常，日志中会出现 Stream 已断开，10s 后重连...；程序会自动重连，无需重启。

💬 iClaw智能聊天与自然语言指令交易

• 智能聊天（多模型）：发送 @模型名 你的问题，机器人会调用对应大模型回复（只聊天不下单）。
  • 示例：@幻银超i 今天市场怎么看？、@deepseek 解释一下市盈率
• 自然语言指令AI交易：发送 @iClaw 你的自然语言交易意图，机器人会调用你文本中提到的模型把意图翻译为 JSON 指令，经校验后复用 /buy、/sell 链路执行。
  • 示例：@iClaw 用幻银超i买入002961 1000股，价格26.9
• /开头的指令交易：
  • 买入/卖出：/buy [模型名] 股票代码 数量 [价格]、/sell [模型名] 股票代码 数量 [价格]，模型名可省略。示例：/buy 000001 1000、/sell 幻银超i 000001 500 27.0
  • 批量买入：/batch_buy [模型名] 股票代码:数量,股票代码:数量,...，模型名可省略（省略则用默认模型）。示例：/batch_buy 000001:1000,000002:500 或 /batch_buy 幻银超i 000001:1000,000002:500
  • 条件交易：/condition [模型名] 股票代码 条件 买入/卖出 [数量]，模型名可省略。条件支持 price>数字、price<数字、price_between 低 高，示例：/condition 000001 price>11 buy 1000 或 /condition 幻银超i 000001 price>11 buy 1000；数量可省略：不写时买入按风险与可用资金自动算数量、卖出为全部持仓；若当下价格未满足条件，系统会在后台按交易轮次间隔持续监控，触发即按市价下单（默认有效期6小时）。条件单监控在独立任务中运行，不占用主程序每轮交易循环。创建后机器人会返回条件单ID，撤销：/cancel <条件单ID>（如 /cancel COND-1734567890-000001）。
• 完整指令列表：在钉钉中发送 /help 或 /h 可查看完整指令列表与示例。
• 远程启动本地应用：发送 /runapp <应用别名或内置名称>，由主程序在本机尝试启动白名单中的桌面应用。
  • 别名通过 user_config.json 的 APP_LAUNCH_CONFIG.aliases 配置，例如 "mytrader": "C:\\Users\\you\\Desktop\\MyTrader\\MyTrader.exe"
  • 对于 notepad、calc 这类规范化程序，可直接 /runapp notepad 使用（也可通过 builtin_allow 扩展）

📊 Ai炒股大赛排行榜、实时收益及决策日志、实时投资组合

安装及启动
项目已打包成windows桌面exe，可直接运行，获取方法：前往https://hyqibot.com/card-shop.html 捐赠相应期限的服务卡即可获得，如需接入实盘需拥有合规通道并自行完成或有要求的相关报备。有疑问可发送您的github地址（必需）+简短介绍到：[email protected]，更多信息，可点：https://www.hyqibot.com/ 有永久免token费的龙虾iclaw赠送。

启动方法很简单：傻瓜式，无需python基础
下载Releases安装包，解压，将user_config.json.example重命名为user_config.json，填写参数
需要用到的参数有几类：1）真实交易账户需合规接入，比如机构自有通道或各大券商QMT；2）自选股票池通达信本地路径；3）deepseek等大模型api-key；4）licence：hyqibot
填好参数保存，右键以管理员身份运行解压好的exe文件，图形界面上点击启动交易即可。

核心飞跃：真正的AI自我发现

1. 从空白或使用者设定的风格开始
   所有AI初始可以都是"自适应投资者"

   也可以预设投资风格

   通过实践发现自己的特长

2. 自我反思机制
   定期回顾投资表现

   分析什么策略对自己有效

   主动调整投资哲学

3. 风格进化
   记录风格变化轨迹

   基于表现调整风险偏好

   逐步形成稳定的投资人格

4. 真正的差异性
   现在4个AI将：

   自主发现风格：比如，幻银超i最明白博弈精髓，DeepSeek可能发现自己擅长量化，Qwen或者偏好价值投资

   不同的进化路径：每个AI基于自身特质和市场反馈形成独特风格

   动态调整：根据市场环境变化调整策略

   真正的个性：发展出反映其"性格"的投资哲学

   预期结果
   您将看到：

   真正的AI个性：每个AI发展出独特的投资身份

   动态的风格进化：投资风格会随着经验积累而改变

   基于表现的调整：成功的策略被保留，失败的被淘汰

   不可预测的结果：无法预先知道哪个AI会发展出什么风格

   这样设计后，您将见证真正的"AI投资经理成长史"，而不是预先编排的表演。每个AI都有机会找到最适合自己的投资道路！

🧠 幻银超i（HYQi）的愿景

我们正以实时的真实交易环境去训练Ai，以此来测试并提高Ai的智慧，运用开放式学习、大规模强化学习等技术来驾驭其复杂性。

心怀此念，我们正广纳贤才，招募资深ai研究员、创业伙伴与思维破局者。

如果您渴望为现实世界打造Alpha-HYQi，欢迎与我们携手。简历请发送至：[email protected]

"资本配置，是ai智慧接受真理检验的最佳试金石"。

———幻银超i与您共勉

📞 支持与社区

💬 讨论交流: GitHub Discussions

🐛 问题反馈: GitHub Issues

📜 许可证条款

1. 开源核心模块（MIT 许可证）

开源部分源代码（位于本仓库 ./codes目录下）基于 MIT 许可证 授权，你享有以下权利：

自由使用、复制、修改、合并、发布、分发、再授权及出售开源核心代码的副本；

将开源核心代码集成到个人 / 商业项目（包括闭源项目），但需保留原始版权声明和 MIT 许可证文本。

完整许可证文本：LICENSE

2. 商业闭源模块
   商业模块（包括但不限于打包成exe文件内嵌的所有模块）受版权保护，基于 商业许可证 提供，核心条款如下：

使用需获得有效商业授权（联系我们获取订阅 / 授权方案）；

禁止未经授权分发、反编译、逆向工程或转售商业模块；

商业模块仅可与本项目的开源核心模块搭配使用，不可单独使用或集成到其他未授权项目。

完整商业许可证详情：LICENSE-COMMERCIAL（购买授权后提供）

🤝 贡献指南

我们欢迎对开源核心模块的贡献！贡献步骤：

Fork 本仓库；

仅在 ./codes 目录下修改代码；

提交 Pull Request，并清晰描述修改内容。

详细贡献规则：CONTRIBUTING.md

注意：商业模块为闭源性质，不接受外部代码贡献。

📞 联系与支持

开源核心用户

问题反馈：通过 GitHub Issues 提交开源核心模块相关的 Bug 报告或功能需求；

社区交流：加入我们的 [GitHub Discussions] 参与讨论。

商业授权用户

技术支持：通过 [[email protected]] 联系专属支持团队（24 小时内响应）；

授权咨询：发送邮件至 [[email protected]]，了解定价、订阅方案或定制授权服务。

⚠️ 免责声明

开源核心模块基于 MIT 许可证以 “原样” 提供，不提供任何明示或暗示的担保（详见完整 MIT 许可证文本）；

商业模块将按照许可证协议提供技术支持，但不保证与所有自定义环境兼容；

双许可证模式可能会调整，任何更新将在本 README 中公示，并通知商业授权用户。

© [2025] 幻银超i 保留所有权利。开源核心模块基于 MIT 许可证授权。商业模块基于商业许可证授权。

A-share-Ai: Phantom Silver Quantitative AI Trading System for A-Shares

An automated A-share trading system powered by four major domestic large language models (HYQi Phantom Silver Ultra-i, DeepSeek, Qwen, Doubao), supporting full-market stock selection and live trading.

Integrates our in-house lobster iClaw: control all four models' fully automated trading from your phone with a single sentence.

Trade Execution

• Buy: Size by total assets × position ratio, and cap spend at a configured share of available cash per cycle.
• Sell (model auto SELL): Sell the entire current position. Not limited by available-cash caps (so sells are not sized to 0 shares when cash is tight).

Key Features

🎯 Let AI showcase its capabilities in the A-share market—multiple large language models make fully autonomous decisions and compete against each other across all A-share stocks!

🎯 Connect to real A-share trading accounts, display the AI's real trading process in real time. Real-time + real trading expands the system's application scenarios.

🎯 Multi-model Integration: Supports HYQi Phantom Silver Ultra-i, DeepSeek, Qwen, Doubao, and other large language models.

🎯 Strategy Evolution: Based on trading performance, AI models continuously optimize their investment strategies.

⏰ Second-level Live Trading Support - Connect to live trading with second-level precision. A compliant trading channel is required.

🚀 Service Deployment & Parallel Execution - Deploy production services + parallel model execution.

🎨 AI Trading Logic Visualization - Detailed trading log visualization (full trading process display).

🎯 Core Feature: 100% AI autonomous decision-making with zero human intervention. AI strategies self-discover, self-train, and self-evolve.

🎯 Special Function: Optional real-time human intervention and interaction with AI during trading. Integrates our in-house lobster iClaw to take over that interaction and command all four AIs at once.

🎯 Newly added asynchronous execution to improve system speed.

🤖 4-Model Parallelism: HYQi Phantom Silver Ultra-i, DeepSeek, Qwen, and Doubao run together to compare returns
📈 Full-Market Stock Selection: AI independently selects promising stocks from the entire market, with customizable stock pools (optional)
💹 Real-Time Market Data: Integrates real-time A-share data for AI use and live trading
💹 Historical Market + Financial Data: Integrates A-share historical market and financial data for AI analysis and training
🎯 Intelligent Trading: Generates buy/sell signals based on AI analysis and executes trades fully automatically
🌈 Multiple Strategies: Includes trend following, moving average strategies, technical indicator strategies, quantitative arbitrage strategies, and more—fully selected and continuously evolved by AI
📊 Real-Time Monitoring: UI interface displays real-time profit curves and trading logs
🔒 Risk Control: Built-in position control and stop-loss mechanisms

📱 iClaw DingTalk Setup

Method 1 (deprecated): After enabling config/dingtalk.json, the app asked you to send /status once in a DingTalk private chat to bind. On success it wrote your account ID to default_conversation_id, then pushed each round's results to that chat (no manual personal ID).

Method 2: Obtain sender_id via a test script or the DingTalk console and write it to default_conversation_id in config/dingtalk.json, along with the other required fields (see the example file). After launch, each round's results are pushed to that account, and DingTalk can remotely control all models in the project in real time.

Stream connection: Inbound messages use WebSocket Stream; outbound notices use the REST API (independent). If private-chat commands stop arriving for a long time while trade notices still work, logs will show Stream 已断开，10s 后重连.... The app reconnects automatically; no restart is required.

💬 iClaw Chat and Natural-Language Trade Commands

• Chat (multi-model): Send @model_name your question. The bot calls that model and replies (chat only, no orders).
  • Examples: @幻银超i 今天市场怎么看？, @deepseek 解释一下市盈率
• Natural-language AI trading: Send @iClaw your natural-language trade intent. The bot uses the model named in your text to translate the intent into a JSON command, validates it, then runs the same /buy / /sell path.
  • Example: @iClaw 用幻银超i买入002961 1000股，价格26.9
• Slash commands:
  • Buy/sell: /buy [model] symbol qty [price], /sell [model] symbol qty [price]. Model name is optional. Examples: /buy 000001 1000, /sell 幻银超i 000001 500 27.0
  • Batch buy: /batch_buy [model] symbol:qty,symbol:qty,.... Model name is optional (default model if omitted). Examples: /batch_buy 000001:1000,000002:500 or /batch_buy 幻银超i 000001:1000,000002:500
  • Conditional orders: /condition [model] symbol condition buy/sell [qty]. Model name is optional. Conditions: price>n, price<n, price_between low high. Examples: /condition 000001 price>11 buy 1000 or /condition 幻银超i 000001 price>11 buy 1000. Qty is optional: if omitted, buys are sized from risk and available cash, sells are the full position. If the price is not met yet, the system keeps watching in the background on the trading-cycle interval and submits a market order when triggered (default TTL 6 hours). Conditional monitoring runs in a separate task and does not block the main trading loop. The bot returns a condition-order ID; cancel with /cancel <id> (e.g. /cancel COND-1734567890-000001).
• Full command list: Send /help or /h in DingTalk.
• Launch a local app remotely: Send /runapp <alias or builtin name>. The host tries to start a whitelisted desktop app.
  • Aliases are configured in user_config.json under APP_LAUNCH_CONFIG.aliases, e.g. "mytrader": "C:\\Users\\you\\Desktop\\MyTrader\\MyTrader.exe"
  • Canonical programs such as notepad and calc can be started with /runapp notepad (extend via builtin_allow)

📊 AI Stock Trading Competition Rankings, Real-Time Returns & Decision Logs, Real-Time Portfolio

Install & Launch
The project ships as a Windows desktop EXE and can be run directly. Get it by donating for a service card of the matching term at https://hyqibot.com/card-shop.html. Live trading requires a compliant channel; complete any required filings yourself. Questions: send your GitHub URL (required) plus a short intro to [email protected]. More info: https://www.hyqibot.com/ . A permanent zero-token-fee lobster iClaw is included as a gift.

Launch is simple: no Python experience required.
Download the Releases package, unzip it, rename user_config.json.example to user_config.json, and fill in the parameters.
Parameter groups: 1) a compliant live-trading hookup, e.g. an institutional channel or broker QMT; 2) local Tongdaxin path for your custom stock pool; 3) API keys for DeepSeek and other models; 4) licence: hyqibot
Save, right-click the unzipped EXE and run as administrator, then click Start Trading in the UI.

Core Leap: True AI Self-Discovery

1. Start from Scratch or User-Defined Style
   All AIs can start as "adaptive investors" initially.
   Investment styles can also be preset.
   Discover their own strengths through practice.
2. Self-Reflection Mechanism
   Regularly review investment performance.
   Analyze which strategies work for themselves.
   Proactively adjust investment philosophy.
3. Style Evolution
   Record the trajectory of style changes.
   Adjust risk appetite based on performance.
   Gradually form a stable investment personality.
4. True Differentiation

Now the 4 AIs will:
Autonomously discover styles: For example, HYQi Phantom Silver Ultra-i masters the essence of trading, DeepSeek may excel in quantitative analysis, and Qwen might prefer value investing.
Different evolutionary paths: Each AI forms a unique style based on its own characteristics and market feedback.
Dynamic adjustment: Adapt strategies to changes in market conditions.
True individuality: Develop an investment philosophy that reflects its "personality."

Expected Outcomes

You will witness:

True AI individuality: Each AI develops a unique investment identity.

Dynamic style evolution: Investment styles change with experience accumulation.

Performance-based adjustments: Successful strategies are retained, failed ones are eliminated.

Unpredictable results: It is impossible to predict which style each AI will develop in advance.

With this design, you will witness a true "AI investment manager growth journey" rather than a pre-scripted performance. Each AI has the opportunity to find the most suitable investment path!

🧠 Vision of HYQi Phantom Silver Ultra-i

We are training AI in a real-time, real trading environment to test and enhance its intelligence, leveraging technologies such as open-ended learning and large-scale reinforcement learning to harness its complexity.

With this vision, we are actively recruiting senior AI researchers, entrepreneurial partners, and visionary thinkers.

If you aspire to build Alpha-HYQi for the real world, we welcome you to join us. Please send your resume to: [email protected]

"Capital allocation is the best touchstone for AI intelligence to be tested by truth."

—— Together with HYQi Phantom Silver Ultra-i

📞 Support & Community

💬 Discussions: GitHub Discussions

🐛 Issues: GitHub Issues

📜 License Terms

1. Open-Core Module (MIT License)

The open-source source code (located in the ./codes directory of this repository) is licensed under the MIT License. You are granted the following rights:

Freely use, copy, modify, merge, publish, distribute, sublicense, and sell copies of the open-core code.

Integrate the open-core code into personal/commercial projects (including closed-source projects), provided that the original copyright notice and MIT License text are retained.

Full license text: LICENSE

2. Commercial Closed-Source Modules

Commercial modules (including but not limited to all modules embedded in the packaged EXE file) are protected by copyright and provided under a Commercial License. Key terms are as follows:

Usage requires a valid commercial license (contact us for subscription/authorization plans).

Unauthorized distribution, decompilation, reverse engineering, or resale of commercial modules is prohibited.

Commercial modules may only be used in conjunction with the open-core module of this project, and shall not be used independently or integrated into other unauthorized projects.

Full commercial license details: LICENSE-COMMERCIAL (provided upon license purchase)

🤝 Contributing Guidelines

We welcome contributions to the open-core module! Contribution steps:

Fork this repository.

Modify code only in the ./codes directory.

Submit a Pull Request with a clear description of changes.

Detailed contribution rules: CONTRIBUTING.md

Note: Commercial modules are closed-source and do not accept external code contributions.

📞 Contact & Support

For Open-Core Users
Issue Reporting: Submit bug reports or feature requests related to the open-core module via GitHub Issues.

Community Discussion: Join our [GitHub Discussions] to participate in discussions.

For Commercial License Users

Technical Support: Contact the dedicated support team via [[email protected]] (response within 24 hours).

License Inquiries: Send an email to [[email protected]] to learn about pricing, subscription plans, or custom authorization services.

⚠️ Disclaimer

The open-core module is provided "as is" under the MIT License, without any express or implied warranties (see the full MIT License for details).

Commercial modules will be provided with technical support in accordance with the license agreement, but compatibility with all custom environments is not guaranteed.

The dual-license model is subject to change. Any updates will be announced in this README and notified to commercial license holders.

© [2025] HYQi Phantom Silver Ultra-i. All rights reserved. Open-Core Module Licensed under MIT License. Commercial Modules Licensed under Commercial License.