MandoCode launches as a local-first AI coding assistant for .NET developers, requiring no API keys and running entirely on your machine or via Ollama.

What happened: MandoCode is a new AI coding assistant built on .NET 8, Semantic Kernel, and Ollama that runs locally or connects to Ollama cloud. It offers project-aware features including reading and writing files across your codebase, web search via DuckDuckGo (or an optional free Tavily key), task planning, and support for file types including C#, JavaScript, TypeScript, Python, CSS, HTML, and JSON — all within the terminal.

Why it matters: Unlike cloud-dependent coding assistants, MandoCode requires no API keys and lets developers keep their code local, addressing privacy and cost concerns. The tool includes safety features like diff approvals that require you to review every file change before it touches disk, giving you full control over what the AI modifies.

What to watch: Setup requires .NET 8 SDK and Ollama, with a guided wizard on first run. Local users should monitor the context window (the amount of conversation and code the model can see at once) — it defaults to ~4k tokens and can cause the model to lose instructions if too low — and the max token cap (which defaults to 32k). The tool is available now via `dotnet tool install -g MandoCode` or from source on GitHub.