Installing vmblu
vmblu is a toolkit for building node- and message-based applications that can be inspected, generated, run, and exposed to agents through controlled capabilities.
vmblu is agent-independent: it does not require a specific coding agent, LLM provider, agent SDK or orchestration framework. At the same time, vmblu is strongly agent-oriented. It is designed for modern software workflows where agents help write code, applications expose standard agent-facing interfaces, and runtime guardrails make those applications safer to operate.
You normally install three things:
- the VS Code extension to inspect and edit vmblu models;
- the npm packages to create, generate and run vmblu applications;
- agent support when you want a coding agent to understand the vmblu workflow.
Requirements
- Node.js 18 or newer.
- Visual Studio Code if you want to use the vmblu editor.
- A JavaScript or TypeScript project target.
- A coding agent such as Codex or Claude if you want agent-assisted project creation and development.
Generated vmblu applications can run in browser or Node.js environments, depending on the runtime selected in the model.
Installation
Ask your agent
Let your agent read this file and ask it to set up the project, or do it yourself:
Install The Editor
Install the vmblu extension from the VS Code Marketplace:
https://marketplace.visualstudio.com/items?itemName=vizualmodel.vmbluIn VS Code:
- Open the Extensions view with
Ctrl+Shift+X. - Search for
vmblu. - Install the extension.
Install The npm Packages
In a project that uses vmblu directly, install the CLI and runtime:
npm install @vizualmodel/vmblu-cli @vizualmodel/vmblu-runtimeThe CLI also installs @vizualmodel/vmblu-core as a dependency. The core package contains the shared model loading and compilation code used by vmblu tooling.
For one-off CLI usage:
npx @vizualmodel/vmblu-cli --helpThe examples below use the short vmblu command. If vmblu is not installed as a command in your project or shell, use npx @vizualmodel/vmblu-cli instead.
Install Agent Support
Agent support installs guidance files so a coding agent can work with vmblu projects more reliably. vmblu is agent-independent, so this support is agent-specific guidance over the same vmblu CLI, model and runtime architecture.
Install Codex support:
vmblu agent install codexInstall Claude support:
vmblu agent install claudeYou can get the list of supported agents as follows.
vmblu agent listWhen the support files are installed, tell the agent:
use vmbluThe agent should then follow the vmblu workflow for the project.
Create A Project With An Agent
The most natural way to start a vmblu project is to tell your coding agent:
use vmbluor give it a more explicit instruction:
Create a new vmblu application called my-new-application.With vmblu support installed, the agent should:
- install or verify the vmblu CLI and runtime packages;
- initialize the project with
vmblu initif needed; - read
.vmblu/vmblu.prompt.md; - use the root
.bluentrypoint to find the model inmodel/; - treat
nodes/as the model-owned source code folder; - generate the profile, application and capabilities when appropriate.
Create A Project Manually
You can also create the project yourself:
vmblu init my-new-applicationThe generated project uses a root .blu entrypoint and keeps model files under model/:
my-new-application/
my-new-application.blu
package.json
model/
my-new-application.mod.blu
my-new-application.mod.viz
nodes/
.vmblu/
vmblu.prompt.md
overrides/
cache/
logs/The root entrypoint points to the actual model file:
{
"kind": "vmblu.entrypoint",
"version": 1,
"model": "model/my-new-application.mod.blu"
}Open the root .blu file in the editor.
Generate And Run
The runtime is selected in the vmblu model, normally through the editor. Users should not edit generated application files to change runtime imports.
The available runtime choices are:
@vizualmodel/vmblu-runtime/rt-base
@vizualmodel/vmblu-runtime/rt-als
@vizualmodel/vmblu-runtime/rt-browser-agent
@vizualmodel/vmblu-runtime/rt-nodejs-agentThe selected runtime determines which model-wide and node-level runtime settings are meaningful. The model stores those settings; the runtime normalizes and enforces them when the application runs.
A coding agent can usually choose the right dev or build command for the selected environment, such as browser or Node.js. See the User Guide for manual workflows.
Try The Examples
The examples live in a separate repository:
https://github.com/vizualmodel/vmblu-examplesVersion Status
vmblu is pre-1.0. The editor, CLI, core package, runtime, and agent integration are usable, but project layout, package boundaries and distribution details may still evolve while the public API is being stabilized.
Feedback
- Issues:
https://github.com/vizualmodel/vmblu/issues - Discussions:
https://github.com/vizualmodel/vmblu/discussions - Website:
https://vmblu.dev