Chat with Your Codebase

A streamlit application for interactions with your codebase & fine-tuning LLM behavior.

• Explore repositories visually and enrich your LLM-queries with context-aware code snippets by using Retrieval Augmented Generation (RAG).
• Save LLM-responses from the Streamlit App directly to your filesystem
• Keep your conversations tidy with a compact, expandable history
• Flexibly finetune LLM-behavior for brainstorming, generation of markdown learning material (markdown & Jupyter notebook) & explanations directly in your browser.

Development Setup

• uv sync creates the virtual environment.
• Export your OPENAI_API_KEY (e.g. in ~/.bashrc).
• Use ./run.sh to start the application

Chat Interface

The sidebar allows you to select pre-defined system prompts and swap models on the fly, giving you fine-grained control over the assistant.

• Chat history appears as expandable QA pairs for easy navigation.
• Save any response as Markdown straight into your Obsidian vault.
• Reset the conversation or pick from themed prompts whenever you need a fresh start.

Codebase Explorer

• A tokenizer walks every GitHub repository in ~ (respecting .gitignore), splitting it into functions and classes stored in a DataFrame.
• This data powers an interactive graph where each node represents a code chunk - node size scales with lines of code.
• "Chat with your codebase" uses Retrieval Augmented Generation to answer with code snippets relevant to your questions.
• Switch between hierarchical or community layouts to spot call relations and module clusters at a glance.