deploy: c65ef92

Author: machineily

README.md

@@ -102,7 +102,7 @@ Modern Flask-based web application with:
 ### 1. Create and Activate Conda Environment
 
 ```bash
-conda create -n quantagents python=3.10
+conda create -n quantagents python=3.11
 conda activate quantagents
 ```
 
@@ -121,13 +121,22 @@ conda install -c conda-forge ta-lib
 
 Or visit the [TA-Lib Python repository](https://github.com/ta-lib/ta-lib-python) for detailed installation instructions.
 
-### 3. Set Up OpenAI API Key
+### 3. Set Up LLM API Key
 You can set it in our Web InterFace Later,
+
 ![alt text](assets/apibox.png)
 
 Or set it as an environment variable:
 ```bash
-export OPENAI_API_KEY="your_api_key_here"
+# For OpenAI
+export OPENAI_API_KEY="your_openai_api_key_here"
+
+# For Anthropic (Claude)
+export ANTHROPIC_API_KEY="your_anthropic_api_key_here"
+
+# For Qwen (DashScope, based in Singapore — delays may occur)
+export DASHSCOPE_API_KEY="your_dashscope_api_key_here"
+
 ```
 
 
@@ -191,24 +200,30 @@ print(final_state.get("pattern_report"))
 print(final_state.get("trend_report"))
 ```
 
-You can also adjust the default configuration to set your own choice of LLMs, analysis parameters, etc.
+You can also adjust the default configuration to set your own choice of LLMs or analysis parameters in web_interface.py.
 
 ```python
-from trading_graph import TradingGraph
-from default_config import DEFAULT_CONFIG
-
-# Create a custom config
-config = DEFAULT_CONFIG.copy()
-config["agent_llm_model"] = "gpt-4o-mini"  # Use a different model for agents
-config["graph_llm_model"] = "gpt-4o"       # Use a different model for graph logic
-config["agent_llm_temperature"] = 0.2      # Adjust creativity level for agents
-config["graph_llm_temperature"] = 0.1      # Adjust creativity level for graph logic
-
-# Initialize with custom config
-trading_graph = TradingGraph(config=config)
-
-# Run analysis with custom configuration
-final_state = trading_graph.graph.invoke(initial_state)
+if provider == "anthropic":
+    # Set default Claude models if not already set to Anthropic models
+    if not analyzer.config["agent_llm_model"].startswith("claude"):
+        analyzer.config["agent_llm_model"] = "claude-haiku-4-5-20251001"
+    if not analyzer.config["graph_llm_model"].startswith("claude"):
+        analyzer.config["graph_llm_model"] = "claude-haiku-4-5-20251001"
+
+elif provider == "qwen":
+    # Set default Qwen models if not already set to Qwen models
+    if not analyzer.config["agent_llm_model"].startswith("qwen"):
+        analyzer.config["agent_llm_model"] = "qwen3-max"
+    if not analyzer.config["graph_llm_model"].startswith("qwen"):
+        analyzer.config["graph_llm_model"] = "qwen3-vl-plus"
+    
+else:
+    # Set default OpenAI models if not already set to OpenAI models
+    if analyzer.config["agent_llm_model"].startswith(("claude", "qwen")):
+        analyzer.config["agent_llm_model"] = "gpt-4o-mini"
+    if analyzer.config["graph_llm_model"].startswith(("claude", "qwen")):
+        analyzer.config["graph_llm_model"] = "gpt-4o"
+        
 ```
 
 For live data, we recommend using the web interface as it provides access to real-time market data through yfinance. The system automatically fetches the most recent 30 candlesticks for optimal LLM analysis accuracy.
@@ -251,8 +266,16 @@ This project is licensed under the MIT License - see the LICENSE file for detail
 
 ## 🙏 Acknowledgements
 
-This repository was built off of [**LangGraph**](https://github.com/langchain-ai/langgraph), [**OpenAI**](https://github.com/openai/openai-python), [**yfinance**](https://github.com/ranaroussi/yfinance), [**Flask**](https://github.com/pallets/flask), [**TechnicalAnalysisAutomation**](https://github.com/neurotrader888/TechnicalAnalysisAutomation/tree/main) and [**tvdatafeed**](https://github.com/rongardF/tvdatafeed).
+This repository was built with the help of the following libraries and frameworks:
 
+- [**LangGraph**](https://github.com/langchain-ai/langgraph)
+- [**OpenAI**](https://github.com/openai/openai-python)
+- [**Anthropic (Claude)**](https://github.com/anthropics/anthropic-sdk-python)
+- [**Qwen**](https://github.com/QwenLM/Qwen)
+- [**yfinance**](https://github.com/ranaroussi/yfinance)
+- [**Flask**](https://github.com/pallets/flask)
+- [**TechnicalAnalysisAutomation**](https://github.com/neurotrader888/TechnicalAnalysisAutomation/tree/main)
+- [**tvdatafeed**](https://github.com/rongardF/tvdatafeed)
 ## ⚠️ Disclaimer
 
 This software is for educational and research purposes only. It is not intended to provide financial advice. Always do your own research and consider consulting with a financial advisor before making investment decisions.
@@ -263,7 +286,7 @@ This software is for educational and research purposes only. It is not intended
 
 1. **TA-Lib Installation**: If you encounter TA-Lib installation issues, refer to the [official repository](https://github.com/ta-lib/ta-lib-python) for platform-specific instructions.
 
-2. **OpenAI API Key**: Ensure your API key is properly set in the environment or through the web interface.
+2. **LLM API Key**: Ensure your API key is properly set in the environment or through the web interface.
 
 3. **Data Fetching**: The system uses Yahoo Finance for data. Some symbols might not be available or have limited historical data.
 
@@ -272,10 +295,12 @@ This software is for educational and research purposes only. It is not intended
 ### Support
 
 If you encounter any issues, please:
+
+0. Try refresh and re-enter LLM API key
 1. Check the troubleshooting section above
 2. Review the error messages in the console
 3. Ensure all dependencies are properly installed
-4. Verify your OpenAI API key is valid and has sufficient credits
+4. Verify your API key is valid and has sufficient credits
 
 ## 📧 Contact
 

README_CN.md

@@ -99,7 +99,7 @@
 ### 1. 创建并激活 Conda 环境
 
 ```bash
-conda create -n quantagents python=3.10
+conda create -n quantagents python=3.11
 conda activate quantagents
 ```
 
@@ -117,19 +117,26 @@ conda install -c conda-forge ta-lib
 
 或访问 [TA-Lib Python 仓库](https://github.com/ta-lib/ta-lib-python) 获取详细的安装说明。
 
-### 3. 设置 OpenAI API 密钥
+### 3. 设置 LLM API 密钥
 您可以在我们的网络界面中稍后设置它，
-![API 密钥设置](assets/apibox.png)
+
+![alt text](assets/apibox.png)
 
 或将其设置为环境变量：
 ```bash
-export OPENAI_API_KEY="your_api_key_here"
+# For OpenAI
+export OPENAI_API_KEY="your_openai_api_key_here"
+
+# For Anthropic (Claude)
+export ANTHROPIC_API_KEY="your_anthropic_api_key_here"
+
+# For Qwen (DashScope, based in Singapore — delays may occur)
+export DASHSCOPE_API_KEY="your_dashscope_api_key_here"
+
 ```
 
 ## 🔧 实现细节
 
-我们使用 LangGraph 构建 QuantAgents 以确保灵活性和模块化。我们使用 gpt-4o 和 gpt-4o-mini 作为我们的深度思考和快速思考 LLM 进行实验。但是，出于测试目的，我们建议您使用 gpt-4o-mini 来节省成本，因为我们的框架会进行大量 API 调用。
-
 **重要说明**：我们的模型需要一个可以接受图像输入的 LLM，因为我们的智能体会生成和分析视觉图表以进行模式识别和趋势分析。
 
 ### Python 使用
@@ -161,24 +168,30 @@ print(final_state.get("pattern_report"))
 print(final_state.get("trend_report"))
 ```
 
-您还可以调整默认配置以设置您自己的 LLM 选择、分析参数等。
+您还可以调整默认配置以在 web_interface.py 中设置您自己的 LLM 选择或分析参数。
 
 ```python
-from trading_graph import TradingGraph
-from default_config import DEFAULT_CONFIG
-
-# 创建自定义配置
-config = DEFAULT_CONFIG.copy()
-config["agent_llm_model"] = "gpt-4o-mini"  # 为智能体使用不同的模型
-config["graph_llm_model"] = "gpt-4o"       # 为图逻辑使用不同的模型
-config["agent_llm_temperature"] = 0.2      # 调整智能体的创造力水平
-config["graph_llm_temperature"] = 0.1      # 调整图逻辑的创造力水平
-
-# 使用自定义配置初始化
-trading_graph = TradingGraph(config=config)
-
-# 使用自定义配置运行分析
-final_state = trading_graph.graph.invoke(initial_state)
+if provider == "anthropic":
+    # Set default Claude models if not already set to Anthropic models
+    if not analyzer.config["agent_llm_model"].startswith("claude"):
+        analyzer.config["agent_llm_model"] = "claude-haiku-4-5-20251001"
+    if not analyzer.config["graph_llm_model"].startswith("claude"):
+        analyzer.config["graph_llm_model"] = "claude-haiku-4-5-20251001"
+
+elif provider == "qwen":
+    # Set default Qwen models if not already set to Qwen models
+    if not analyzer.config["agent_llm_model"].startswith("qwen"):
+        analyzer.config["agent_llm_model"] = "qwen3-max"
+    if not analyzer.config["graph_llm_model"].startswith("qwen"):
+        analyzer.config["graph_llm_model"] = "qwen3-vl-plus"
+    
+else:
+    # Set default OpenAI models if not already set to OpenAI models
+    if analyzer.config["agent_llm_model"].startswith(("claude", "qwen")):
+        analyzer.config["agent_llm_model"] = "gpt-4o-mini"
+    if analyzer.config["graph_llm_model"].startswith(("claude", "qwen")):
+        analyzer.config["graph_llm_model"] = "gpt-4o"
+        
 ```
 
 对于实时数据，我们建议使用网络界面，因为它通过 yfinance 提供对实时市场数据的访问。系统会自动获取最近 30 个蜡烛图以获得最佳的 LLM 分析准确性。
@@ -213,7 +226,7 @@ python web_interface.py
 2. **时间框架选择**：分析从 1 分钟到每日间隔的数据
 3. **日期范围**：为分析选择自定义日期范围
 4. **实时分析**：获得带有可视化的全面技术分析
-5. **API 密钥管理**：通过界面更新您的 OpenAI API 密钥
+5. **API 密钥管理**：通过界面更新您的 LLM API 密钥
 
 ## 📺 演示
 
@@ -244,7 +257,16 @@ python web_interface.py
 
 ## 🙏 致谢
 
-此仓库基于 [**LangGraph**](https://github.com/langchain-ai/langgraph)、[**OpenAI**](https://github.com/openai/openai-python)、[**yfinance**](https://github.com/ranaroussi/yfinance)、[**Flask**](https://github.com/pallets/flask)、[**TechnicalAnalysisAutomation**](https://github.com/neurotrader888/TechnicalAnalysisAutomation/tree/main) 和 [**tvdatafeed**](https://github.com/rongardF/tvdatafeed) 构建。
+此仓库基于以下库和框架构建：
+
+- [**LangGraph**](https://github.com/langchain-ai/langgraph)
+- [**OpenAI**](https://github.com/openai/openai-python)
+- [**Anthropic (Claude)**](https://github.com/anthropics/anthropic-sdk-python)
+- [**Qwen**](https://github.com/QwenLM/Qwen)
+- [**yfinance**](https://github.com/ranaroussi/yfinance)
+- [**Flask**](https://github.com/pallets/flask)
+- [**TechnicalAnalysisAutomation**](https://github.com/neurotrader888/TechnicalAnalysisAutomation/tree/main)
+- [**tvdatafeed**](https://github.com/rongardF/tvdatafeed)
 
 ## ⚠️ 免责声明
 
@@ -256,7 +278,7 @@ python web_interface.py
 
 1. **TA-Lib 安装**：如果您遇到 TA-Lib 安装问题，请参考[官方仓库](https://github.com/ta-lib/ta-lib-python)获取平台特定的说明。
 
-2. **OpenAI API 密钥**：确保您的 API 密钥在环境中或通过网络界面正确设置。
+2. **LLM API 密钥**：确保您的 API 密钥在环境中或通过网络界面正确设置。
 
 3. **数据获取**：系统使用雅虎财经获取数据。某些符号可能不可用或历史数据有限。
 
@@ -265,16 +287,18 @@ python web_interface.py
 ### 支持
 
 如果您遇到任何问题，请：
+
+0. 尝试刷新页面，重新通过页面输入LLM API 密钥
 1. 检查上面的故障排除部分
 2. 查看控制台中的错误消息
 3. 确保所有依赖项都正确安装
-4. 验证您的 OpenAI API 密钥有效且有足够的积分
+4. 验证您的 LLM API 密钥有效且有足够的积分
 
 ## 📧 联系
 
 如有问题、反馈或合作机会，请联系：
 
-**邮箱**：[[email protected]](mailto:[email protected])
+**邮箱**：[[email protected]](mailto:[email protected]), [[email protected]](mailto:[email protected])
 
 ## Star History
 

assets/apibox.png

@@ -1,7 +1,11 @@
 DEFAULT_CONFIG = {
     "agent_llm_model": "gpt-4o-mini",
     "graph_llm_model": "gpt-4o",
+    "agent_llm_provider": "openai",  # "openai", "anthropic", or "qwen"
+    "graph_llm_provider": "openai",  # "openai", "anthropic", or "qwen"
     "agent_llm_temperature": 0.1,
     "graph_llm_temperature": 0.1,
-    "api_key": "",
+    "api_key": "sk-",  # OpenAI API key
+    "anthropic_api_key": "sk-",  # Anthropic API key (optional, can also use ANTHROPIC_API_KEY env var)
+    "qwen_api_key": "sk-",  # Qwen API key (optional, can also use DASHSCOPE_API_KEY env var)
 }

default_config.py

@@ -6,7 +6,7 @@
 import copy
 import json
 
-from langchain_core.messages import ToolMessage
+from langchain_core.messages import ToolMessage, HumanMessage
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 
 
@@ -44,14 +44,18 @@ def indicator_agent_node(state):
         ).partial(kline_data=json.dumps(state["kline_data"], indent=2))
 
         chain = prompt | llm.bind_tools(tools)
-        messages = state["messages"]
+        # messages = state["messages"]
+        messages = state.get("messages", [])
+        if not messages:
+            messages = [HumanMessage(content="Begin indicator analysis.")]
+
 
         # --- Step 1: Ask for tool calls ---
         ai_response = chain.invoke(messages)
         messages.append(ai_response)
-
+        
         # --- Step 2: Collect tool results ---
-        if hasattr(ai_response, "tool_calls"):
+        if hasattr(ai_response, "tool_calls") and ai_response.tool_calls:
             for call in ai_response.tool_calls:
                 tool_name = call["name"]
                 tool_args = call["args"]
@@ -68,11 +72,52 @@ def indicator_agent_node(state):
                 )
 
         # --- Step 3: Re-run the chain with tool results ---
-        final_response = chain.invoke(messages)
+        # Keep invoking until we get a text response (not another tool call)
+        # This is important for Claude which may make multiple tool calls
+        max_iterations = 5  # Prevent infinite loops
+        iteration = 0
+        final_response = None
+        
+        while iteration < max_iterations:
+            iteration += 1
+            final_response = chain.invoke(messages)
+            messages.append(final_response)
+            
+            # If there are no tool calls, we have the final answer
+            if not hasattr(final_response, "tool_calls") or not final_response.tool_calls:
+                break
+            
+            # If there are more tool calls, execute them
+            for call in final_response.tool_calls:
+                tool_name = call["name"]
+                tool_args = call["args"]
+                tool_args["kline_data"] = copy.deepcopy(state["kline_data"])
+                tool_fn = next(t for t in tools if t.name == tool_name)
+                tool_result = tool_fn.invoke(tool_args)
+                messages.append(
+                    ToolMessage(
+                        tool_call_id=call["id"], content=json.dumps(tool_result)
+                    )
+                )
+
+        # Extract content - handle both string and empty content cases
+        if final_response:
+            report_content = final_response.content
+            # If content is empty or None, try to get text from recent messages
+            if not report_content or (isinstance(report_content, str) and not report_content.strip()):
+                # Check if there's any text content in the messages (skip tool calls)
+                for msg in reversed(messages):
+                    if (hasattr(msg, 'content') and msg.content and 
+                        isinstance(msg.content, str) and msg.content.strip() and 
+                        not hasattr(msg, 'tool_calls')):
+                        report_content = msg.content
+                        break
+        else:
+            report_content = "Indicator analysis completed, but no detailed report was generated."
 
         return {
-            "messages": messages + [final_response],
-            "indicator_report": final_response.content,
+            "messages": messages,
+            "indicator_report": report_content if report_content else "Indicator analysis completed.",
         }
 
     return indicator_agent_node