AgentCode is a web-based chat interface with a chatbot that is connected to a vector database of a GitHub repo. This agent leverages data from a GitHub repository to respond intelligently to user queries. It uses GPT from OpenAI, coupled with the Pinecone vector search engine for efficient information retrieval and storage.
- Enter your GitHub, OpenAI, and Pinecone API keys, and your GitHub repository name, in the
.envfile. - Install the required packages. Run
pip install -r requirements.txtfor Python dependencies andnpm installfor JavaScript dependencies. - Initiate the Flask API by running
python app.py. - Start the React application by running
npm start. - Navigate to localhost:3000 in your web browser.
- Enter your query in the input field and click the 'Submit' button to get a response.
The AgentCode application consists of three main components:
Flask API server and a single endpoint, /chat. The /chat endpoint accepts POST requests comprising a prompt field in the request body, signifying the user's question for the chatbot. The endpoint processes the prompt using the conversational agent derived from agent_setup.py and returns the agent's response.
AgentSetup class, responsible for initializing the conversational agent:
- Utilizing the GitHub API to fetch all relevant files from a specific GitHub repository. It focuses on particular file types (e.g., .md, .txt, .py).
- Creating a "knowledge base" by dividing the content of each fetched file into segments of five lines each. These segments are stored with their corresponding filename.
- Initializing the Pinecone vector search engine and storing each segment from the knowledge base in it.
- Setting up the agent with multiple tools, including a language model for conversation and a tool to retrieve information from the Pinecone knowledge base.
On instantiation of the AgentSetup class and calling its run method, these steps are executed, returning the initialized agent.
This script sets up the React frontend, which houses a text input field for user queries, a 'Submit' button to forward the query to the backend, and a text output field for displaying the agent's response.
On clicking the 'Submit' button, the current content of the input field is sent as a POST request to the Flask API's /chat endpoint. The Flask API processes the prompt using the conversational agent and sends back the response, which the frontend then presents in the output field.
Feel free to contribute!
Use RepoToText to use a GitHub repo with Code Interpreter
- Finish connecting front end and back end
- Fix issues on back end (retrieving data and sending it to the front end)
- Adjust UI for a more attractive text output (code)
- Build input field on UI for GitHub repo input
- Build history on UI to switch between repo's
- Create ability to update vector db when repo changes
This project is licensed under the MIT License. Enjoy using AgentCode and let us know of any cool improvements you make.