wtf is hi-code?

Hi Code is a desktop application that gives you a coding agent you run on your own computer instead of through a hosted service. It is built to work with any AI model that speaks the OpenAI-compatible chat completions format, including Chinese providers such as Kimi, DeepSeek, and Qwen, as well as local models run through Ollama or vLLM. Unlike a simple chat window, the agent can read and write files, run commands, and search through a codebase on its own, repeating steps until a task is finished. Every change is shown to you as a colored diff before it happens, and the tool asks for permission before touching your files unless you turn on a fully automatic mode. There are two ways to use it: a full desktop app built with Electron that includes a chat panel, a visual diff viewer, and a Git panel, or a terminal interface for people who prefer working from the command line. The project also includes an experimental multi-agent setup, where different roles such as an architect, a coder, a reviewer, and a tester can work together on the same task, each with different levels of file access. It supports connecting to external tools through the MCP protocol, and it can load project instructions automatically from files like README or AGENTS.md. The README describes the project as being in an early stage, with the core features working but still under active development. It requires Node.js version 18 or newer to build and run. The project is released under the MIT license and is not affiliated with OpenAI or Anthropic, even though it is designed to feel similar to tools like Claude Code or Codex.