The backend contains the Python code for ingesting data from Github Discussions and Issues, writing that data to a vector database, and providing an API for responding to questions via an LLM and context from the database.
The frontend is a simple web interface for asking questions and getting answers from the API.
You can use Codespaces/Devcontainers to run this application. Start
the remote dev container. From there, you can run make dev-api in a terminal
to start the backend, and then open a new termianl and run make dev-web t
to start the front-end.
If you choose to do manually installation, create a virutal environment and install dependencies for both python and node.
pyenv virtualenv 3.10 llm-support-bot
pyenv local llm-support-bot
make install
You will need some API tokens in order to run the bot.
# Only required to fetch new issues/disccusions to seed Vector Database
GITHUB_API_TOKEN: Github API token with read:discussion permission scope
# Required for answering questions by the LLM
OPENAI_API_KEY:
# Name of the model to use, e.g gpt4 or gpt-3.5-16k
OPENAI_MODEL
# API Key to query Pinecone for embeddings
PINECONE_API_KEY
Start the backend:
make dev-api
Start the frontend
make dev-web