Kittle.

The problem

Opening an unfamiliar codebase usually means hours of reading files in the wrong order before the shape of the project becomes clear. I wanted a tool that lets you ask the codebase questions and see its structure as a diagram, instead of reading it from the first file.

What I built

Kittle. is a chat interface that takes a GitHub repository or files from your machine as context and answers questions about them. It connects to several AI providers: Google Gemini 2.0 (including the Thinking model), OpenAI GPT-4o, Anthropic Claude 3.5 and DeepSeek R1, so you can pick the model per task. Its Design Mode turns a conversation into Mermaid.js architecture diagrams, so the structure of a project is something you can look at, not only read about.

Decisions

• No server. Kittle. is a static Next.js app. Every AI call goes straight from the browser to the provider, and your API keys stay in your browser's local storage. Nothing passes through a backend I run, which keeps keys private and hosting free.
• Bring your own model. Supporting four providers behind one interface means the tool keeps working when one model is better, cheaper or unavailable.
• Motion with purpose. The interface uses Framer Motion and a glass style so long answers and diagrams appear in a clear order rather than all at once.

Built with Next.js 16, React 19, TypeScript, Tailwind CSS 4, Mermaid.js and Framer Motion, and deployed automatically with GitHub Actions.

Where it stands

Started in January 2026 and live at kittle.temrevil.com. Today Kittle. works as a chat assistant over your code; the next step is making it agentic, so it can act on a repository, not only describe it. The source is public on GitHub.