Modern LLMs make it easy to “vibe code”. But as your app grows, features break, code bloat creeps in and the big picture fades.
vmblu fixes that by making your architecture structured, visual, and AI-native. With vmblu, you can co-write your project with an LLM while keeping the architecture clear, the codebase maintainable, and the system navigable.
▞ What is vmblu?
vmblu is for developers who understand that AI changes how software is built — and who want architecture that keeps AI correct, safe, and observable.
vmblu is a graphical editor, format spec and tool set, for building software as a network of message-passing nodes. vmblu makes the architecture of your software explicit and easier to navigate and understand — by humans and LLMs.
vmblu is
- AI-native - LLMs can design the architecture, write the code and interact with the running system.
- A visual modeler - makes your architecture explicit and navigable
- A runnable scaffold - models execute via the vmblu runtime
- Framework-agnostic - use any stack or package, from JS/TS to Svelte or Three.js
The vmblu model is not documentation, it is the actual system.
▞ vmblu is for developers
- vmblu is a tool targeted at developers whose role is changing with the arrival of powerful AI.
- vmblu is for building real, complex systems. It is an excellent foundation to build agentic systems.
- Message-based architectures — the foundation of vmblu — bring their own advantages.
- Migrating an application to a vmblu based design is not complicated. The vmblu editor itself is now a vmblu application.
- vmblu is currently at version 0.0.30, with version 1.0.0 planned in the first quarter of the year (2026). It is in good shape, but expect updates.
▞ Quick start
- Install the VS Code extension
- Add the runtime with:
npm install @vizualmodel/vmblu-runtime
- Add the cli with:
npm install @vizualmodel/vmblu-cli
- Read more at 🌐 vmblu.dev
- Contribute on 🔗 GitHub
▞ Open source
vmblu is licensed under the Apache License 2.0.
▞ Screenshot
vmblu
clarity at scale
