Ironsmith: a menu bar app that turns prompts into small native Mac apps
Create personal Mac apps instantly with a prompt. Supports on-device and cloud LLMs
At a glance
- What is it?
- A GPL-3.0 macOS menu bar forge that generates, builds and saves real SwiftUI apps from a prompt, working with local models through Ollama or hosted keys, sandboxed by default and requiring no Xcode.
- Who is it for?
- Ironsmith fits Mac owners on macOS 26 who keep wishing small utilities into existence, want them built locally with Ollama or their own keys, and will keep the default sandbox on for anything they did not write themselves. Skip it on older macOS, if you need cross-platform output, or for anything beyond focused utilities where a real development workflow still wins.
- Can I use it commercially?
- Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
- Is it still maintained?
- Yes. The repository last received commits 6 days ago.
- What is it written in?
- Mainly Swift, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 17, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
A menu bar app that forges menu bar apps
Ironsmith, from Jeidoban, is a free, open-source macOS menu bar application with a single loop: describe a small app in a prompt, and Ironsmith generates a native Swift and SwiftUI application, builds it, runs it, lets you edit it, and exports it to your Applications folder. The examples gallery shows the target scale, a synthesizer, a painting app, a HEIF converter, an SVG editor, a notepad, a network visualizer, and the README describes the sweet spot as focused utilities you wish existed but would not hunt down or build yourself.
With stronger models, the README names GPT-5.6 Sol and Fable 5, it says surprisingly sophisticated apps are possible. The project is GPL-3.0, ships three releases, and its last push was on 2026-09-12, positioning it as a working tool in the newest and noisiest category: personal software generated on demand.
The model matrix: local first
Ironsmith was designed local-AI-first. Ollama works out of the box, and any OpenAI-compatible API fits, which the README names for LM Studio and Llama.cpp users. Hosted models are supported through your own keys for OpenAI, Anthropic and Gemini, and a sign-in path to Ironsmith grants access to them immediately, with existing ChatGPT logins also supported so no new key has to be created for that route.
Generation itself runs through specialized coding agents: Ironsmith's in-house agents handle tiny macOS apps, and OpenAI's Codex can be selected for more complex projects. The local-first matrix is the differentiator worth weighing: an owner of a beefy Mac can generate personal utilities with no API key, no per-app cost and no data leaving the machine, while users of hosted models trade privacy for capability.
Safety defaults that assume the code might be wrong
Generated code is generated code, and Ironsmith's architecture takes that seriously. Every generated app is built as a signed bundle with sandboxing and hardened runtime enabled by default, which the README credits with greatly reducing the impact of bugs, mistakes or malicious behavior. Sensitive permissions such as camera and microphone access must be explicitly enabled rather than arriving silently.
The protections can be disabled, and the README pairs that power with its own warning: if you turn them off, review the code before running it. That is the correct posture for prompt-generated software, and it is stated in the project's own documentation rather than left as a discoverable surprise. The combination, sandbox by default, permissions gated, escape hatch documented, is a small but complete trust model for a category that mostly lacks one.
No Xcode, on purpose
Generated apps are Swift packages built entirely with the lightweight Xcode command line tools, not full Xcode, and the README one-ups that: Ironsmith itself does not use Xcode either. On first launch the app checks for the command line tools, since generated apps compile locally, and prompts through macOS if they are missing; manual installation is the familiar single command:
xcode-select --installRequirements are macOS 26 or newer on Intel or Apple Silicon, with Apple Intelligence enabled where available, because Ironsmith uses it to generate app icons and to power its built-in Foundation Model. The macOS 26 floor is a real gate that excludes older machines, but it is also what makes the icon generation and on-device model features possible without any external service.
Developing Ironsmith itself
The project builds its own dogfood: development requires the same macOS 26 and command line tools, with Xcode explicitly not required, and a script suite handles the lifecycle:
script/build.sh run
script/test.sh
script/clean.shBuild and run, tests, and cleaning of SwiftPM and script outputs each have a script, and a configuration template copies to Config/.env with a developer signing identity to avoid repeated keychain prompts when running fresh builds. Issues and pull requests follow a contributing guide, and the project is present on GitHub Releases plus its own website for downloads.
For anyone studying how prompt-to-app pipelines are built, the repository is a compact reference: prompt handling, agent selection, Swift package generation, command-line compilation, code signing, sandboxing and export, all in one GPL codebase that fits in a single read.
Limits, and the neighbors
The boundaries are clear. The macOS 26 requirement excludes every machine not running the newest release, and Ironsmith is explicitly aimed at small personal apps; complex projects are delegated to Codex, which signals where the in-house agents stop. Generated quality tracks the model you attach, and the sandbox can be disabled by the user, which moves the trust burden back onto code review exactly when review is hardest, when the code was just generated.
The neighboring products are the hosted prompt-to-app services and general coding agents. Hosted services generate cross-platform apps but own the runtime and your descriptions; general agents can build anything but require you to assemble the build-sign-export loop yourself. Ironsmith's niche is the local, private middle: real native Mac apps, forged on your machine, from your model, at the scale where a menu bar popover is the whole interface.
Editorial conclusion
Ironsmith fits Mac owners on macOS 26 who keep wishing small utilities into existence, want them built locally with Ollama or their own keys, and will keep the default sandbox on for anything they did not write themselves. Skip it on older macOS, if you need cross-platform output, or for anything beyond focused utilities where a real development workflow still wins. Verify the loop in ten minutes: install, connect Ollama, prompt the README's screenshot-renamer example, and run the generated app from the menu bar before deciding how much of your utility belt to forge this way.
Frequently asked questions
What is Ironsmith?
A free, open-source macOS menu bar app that generates small native SwiftUI applications from a text prompt: it creates, builds, runs, edits and exports them to your Applications folder, licensed under GPL-3.0.
Does Ironsmith require Xcode?
No. Generated apps are Swift packages built with the lightweight Xcode command line tools, and Ironsmith itself does not use Xcode either. You need macOS 26 or newer and the command line tools installed via xcode-select.
Which AI models does Ironsmith support?
Ollama out of the box locally, any OpenAI-compatible API such as LM Studio or Llama.cpp, and hosted keys for OpenAI, Anthropic and Gemini, plus a ChatGPT login path. In-house agents handle small apps and OpenAI Codex can be chosen for complex projects.
Community notes