Introducing IronCode: An AI Coding Agent Built for the Terminal
IronCode is an open-source AI coding agent that runs in your terminal, reads your codebase, and ships production-ready code — without uploading your files to the cloud.
We built IronCode because existing AI coding tools kept getting in our way.
Most tools are IDE plugins that autocomplete one line at a time. A few are web interfaces where you paste code into a chat box. Almost all of them send your entire codebase to a remote server. And none of them feel like a real collaborator — something you can hand a task to and trust it to finish.
IronCode is different. It's a CLI-first AI coding agent that understands your whole project, executes multi-step tasks autonomously, and keeps your code on your machine.
What IronCode actually does
When you run ironcode, you get a terminal interface where you can describe a task in plain English:
$ ironcode
> Add rate limiting to the /api/auth endpoints using the existing Redis client
IronCode reads the relevant files, writes the code, runs the tests, and commits — without you touching a keyboard again until it's done.
It's not a chatbot. It's an agent. The difference matters: a chatbot waits for you to copy-paste code and tell it what happened next. An agent reads the error, figures out why it happened, fixes it, and tries again.
Built on Bun and Rust
Performance was a first-class concern from the start. IronCode runs on Bun — not Node.js — which means faster startup, faster file I/O, and a native test runner we actually use.
The file-reading, code-indexing, and diff-generation paths are implemented in Rust via FFI. In practice this means 2–10x faster context building compared to a pure TypeScript implementation — which matters when your codebase has 50,000 lines and the agent is making dozens of tool calls per task.
Privacy by default
Every AI coding tool faces the same question: what happens to your code?
IronCode's answer: only prompts and responses leave your machine. The agent reads your files locally, constructs a prompt, sends it to the AI provider of your choice (OpenAI, Anthropic, or any OpenAI-compatible endpoint), and receives a response. Your source files never leave your filesystem.
You choose the model. You choose the provider. You can run it against a local model via Ollama if you want zero external network calls.
Skills: composable workflows
One of the things that makes IronCode more than a general-purpose agent is the skills system. Skills are pre-built, opinionated workflows for common tasks:
/tdd— writes a failing test first, then the minimum code to pass, then refactors/qa— tests your app in a headless browser, documents every bug with screenshots/code-ship— merges, runs tests, bumps the version, writes the changelog, opens the PR/debug— root cause first, no fixes until the cause is understood
Skills encode good engineering practices. You can read them, fork them, and write your own.
Open source, MIT licensed
IronCode is fully open source under the MIT license. The entire codebase is on GitHub — including the Rust native modules, the skills library, and the CLI.
npm install -g ironcode-ai
ironcode
If you run into bugs or have ideas, open an issue. PRs are welcome.