Chat with Code

Local chat with your local git based repository, built using Ollama, HuggingFaceEmbeddings, LangChain.

Introduction

This project provides ready to use chat with already codebase that was locally on machine, so it is readt to run and use. To use it please update knowledge base with your repo then run it again.

Why this project?

I needed to have ability to chat with my private repo (suitable for company that seek privacy) so i developed this app.

Demo

As example i used next repo as knowledge base and result :

How to use

• install Ollama
• pull model deepseek-coder-v2 using ollama pull deepseek-coder-v2:16b
• install python 3.11
• run pip3 install -r requirements.txt
• run python3.11 -m streamlit run app.py

Update Knowledge base

To update KB we need to chunck codebase and make embededings then save to FAISS

you need to add next snippits(let's assume codebase is JS based project):

# path to local git repo
repo_path = "./project/"
repo = Repo(repo_path)

# Extract text from the repository
file_texts = []
for file in repo.tree().traverse():
    if file.type == "blob":
        if file.name.endswith((".js", ".tsx", ".jsx")):  # Handle JavaScript files
            file_content = file.data_stream.read().decode("utf-8")

            js_splitter = RecursiveCharacterTextSplitter.from_language(
                language=Language.JS, chunk_size=200, chunk_overlap=0
            )
            chunks = js_splitter.create_documents([file_content])
            file_texts.extend([chunk.page_content for chunk in chunks])            


# place this code after embeddings
vectorstore = FAISS.from_texts(file_texts, embeddings)

vectorstore.save_local("vector_store_index")

then when vectorstore is saved you can comment above lines, and use:

vectorstore = FAISS.load_local("vector_store_index",embeddings, allow_dangerous_deserialization=True )

Inspiration

I have inspired from many projects, but i did not get what i want exactly.

• chat_with_github
• QA-Pilot