Chat-GPT clone using LangGraph to implement a Retrieval Augmented Generation (RAG) agent, equipped with web search and document summary tools, wrapped in a Streamlit GUI.
Tools available to RAG Agent:
- Web search
- Chat history
- PDF document summary (coming)
Language: Python 3.10 Models: llama3.2:3b-instruct-fp16, llama3.1
py -m venv \path\to\venv
\path\to\venv\Scripts\activate.bat
pip install -r requirements.txt
Chat Search-and-Summary's RAG agent uses Tavily to search the web. As a result it is necessary to enter a Tavily API key. To do this:
(1) Sign up for a Tavily account and obtain a key (it's free)
(2) Create a secrets.toml in the project directory (should be same level as main.py)
(3) In secrets.toml, write TAVILY_API_KEY = "your-key" and save file.
\path\to\project> streamlit run main.py
- Chat history tool is currently another llm that is queried by the RAG agent with vector embeddings of the chat history as context
(1) Creating vector embeddings with every query is very slow
(2) This llm should at least be another agent and not a tool- or figure out how to have chat history persist outside of streamlit
