We do this by using vector embeddings (https://platform.openai.com/docs/guides/embeddings) to index a specific codebase in a vector database. We then use OpenAI’s GPT 4 model with a dedicated prompt that combines your query and the most relevant results, based on similarity from the vector database, as context to provide a relevant response to your question. The answer is generated as markdown to display things like blocks of code. We also let you see the relevant snippets of code used to generate the answer for reference, ask follow up questions, and we have an option to share an answer.

My co-founder Naren and I used to lead engineering teams. We often had to stop the team from working on their main project and ask them to write documentation to help onboard new engineers to the codebase. That documentation takes days of work to write and ends up being stale very quickly.

Wolfia Codex allows new engineers unfamiliar with a codebase to ask pointed questions without having to always interrupt a co-worker to get answers, nor having to have them write lots of documentation.

We would love to hear what you think: don’t hesitate to share good and bad answers in the comments!