Full-stack developer building tools that make development better. These days that mostly means local AI tooling, .NET, and making a terminal feel like a place you'd actually want to spend the day.

An open-source AI coding assistant that runs on open-weight models, not one company's closed API. It talks to Ollama, so you can run a model entirely on your own machine (free, private, no API key, nothing billed per token), or use a hosted model like kimi-k3:cloud or deepseek-v4.1-flash:cloud when you want more horsepower. Same assistant either way.

It comes in two flavors, and they share the same brain:

Desktop doesn't reimplement anything. It pulls in the CLI's engine as a git submodule, so every improvement to one shows up in the other. Both run on Microsoft Agent Framework and .NET 10.

• Work on real projects. It reads, writes, and searches your code, runs commands, and browses the web, and nothing touches a file until you've seen the diff and said yes.
• Make plans that survive a crash. Bigger jobs become a plan you can review and edit before it starts. If MandoCode closes halfway through, you pick up where it left off instead of starting over.
• Show its work as it goes. You watch replies being written, and a context meter tells you when the conversation is getting too long for the model to remember, so you can /compact it first.
• Run a team of agents (Desktop). Up to four agents side by side, each with its own folder and model. Mention one with @ to hand it a task, and keep chatting while it works.
• Browse and preview (Desktop). A built-in browser for you and your agents, a live view of every command they run, and previews for files, PDFs, and the site you're building.
• Look like whatever you want (Desktop). 25 themes, from Tokyo Night to a period-correct Windows 98 desktop, a flickering CRT, and a Solari split-flap board that can only show uppercase, because real flaps only had one character printed on them.
• Play lofi while it thinks. There's a music player. In a coding assistant. Yes, really.

• v0.15 moved the whole engine off Semantic Kernel and onto Microsoft Agent Framework. It's the biggest change to MandoCode's insides since I wrote it, and it's what made resumable plans possible.
• v0.15.1 made replies stream in live and added the context meter, /history, and scrollable diffs for big changes.

Day-to-day: C#, .NET 10, Blazor, WinUI 3, Microsoft Agent Framework, EF Core, SQL, JS, Python, Node.js

AI stack: Ollama (local + cloud models), open-weight LLMs, tool calling, MCP

Blockchain: Solana ecosystem, Web3.js, on-chain program development

Game dev: Actively exploring game development. Building interactive systems is what got me into programming in the first place.

I treat my editor like a cockpit. On any given day you'll find me jumping between:

• 🖥️ Visual Studio: heavy .NET work, debugging, profiling
• 📝 VS Code: quick edits, web projects, extensions for everything
• 🤖 Claude Code: AI pair programming from the terminal
• 🧠 Codex: rapid prototyping and code generation
• ⚡ MandoCode: because I built my own, so of course I use it. Ollama GO! 🦙

The best tool is the one that fits the problem. I use all of them.

The best developer tools stay out of your way. I built MandoCode because I wanted an AI assistant that:

• Works with any model. Local weights on your GPU or a cloud model through Ollama. Your choice, and you can switch any time.
• Lives where you already work. A terminal app, or a real native Windows app. No browser tab, no Electron.
• Keeps you in charge. It shows you every change before making it, and tells you honestly when something failed instead of pretending it worked.

If that sounds like your kind of thing, pull it down and break it. PRs welcome.

_{Currently vibing to lofi beats piped through a terminal while a cloud model thinks about your code.}