Chatbot Basic

A Streamlit-powered chatbot application that supports both OpenAI GPT and Ollama models, with integrated Arxiv paper retrieval as a tool.

Features

• Conversational AI: Chat with an assistant powered by either OpenAI GPT-4o or Ollama (Llama3.2).
• Arxiv Paper Retrieval: Use the retrieve tool to fetch and summarize academic papers from Arxiv.
• Streaming Responses: AI responses are streamed for a smooth chat experience.
• Model Switching: Subtitle displays which model is currently active.
• LangSmith Tracing: Optional tracing for debugging and analytics.

Setup

1.Clone the repository:

   git clone https://github.com/wangychn/streamlit-chatbot.git
   cd streamlit-chatbot

2.Install dependencies:

   pip install -r requirements.txt

3.Set environment variables:

• Create a .env file in the root directory:

   OPENAI_API_KEY=your-openai-key
   LANGCHAIN_API_KEY=your-langsmith-key

• (Optional) Set additional LangSmith tracing variables if needed.

4.Start Ollama (if using Ollama):

• Make sure Ollama is running locally and the desired model (e.g., llama3.2) is available.

Usage

Run the Streamlit app:

streamlit run src/streamlit_app.py

• The app will display a chat interface.
• The subtitle will show which model is active (GPT-4o or Ollama).
• Type your questions in the chat input.
• To retrieve papers, mention your need to investigate papers in your message.

File Structure

src/
  chatbot.py         # Chatbot logic and graph construction
  streamlit_app.py   # Streamlit UI and app entry point

Customization

• Switch models: Change the llm initialization in streamlit_app.py to use either ChatOpenAI or ChatOllama.
• Add tools: Extend chatbot.py to add more tools or retrieval functions.

Troubleshooting

• Session state errors: Always run the app with streamlit run, not plain python.
• Model errors: Ensure your API keys and Ollama server are set up correctly.
• Streaming issues: Input is disabled while the AI is generating a response to prevent interruptions.