Modern LLMs make it easy to “vibe code”. But as your app grows, features break, code bloat creeps in and the big picture fades. It is hard to find out what exactly is going on in a non-trivial application.
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 a format spec and tool set, for building trusted, maintainable software with AI. vmblu makes the architecture of your software explicit and easier to navigate and understand — by humans and LLMs. With vmblu architecture comes first.
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 - vmblu nodes communicate via messages. The vmblu runtime that takes care of message switching between nodes and blocks unauthorised behaviour.
- 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. The graphical vmblu model lets you understand and reason about the architecture. The vmblu runtime keeps the application safe.
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 — enable this.
vmblu is currently at version 0.x.y, with version 1.0.0 planned in the second 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
