vmblu is a toolchain for building and maintaining software created with coding agents.

vmblu is about architecture: at its core is a blueprint, a structured, machine-readable model that defines nodes, routes, and contracts. The blueprint is the authoritative source for generation, testing, runtime behavior, and agent interaction. Everything else derives from it.

vmblu integrates with your existing stack. You keep your frameworks and libraries. It ships today as a VS Code extension, with other environments to follow.

vmblu models the architecture of a system as a collection of interacting nodes. Nodes communicate through pins with formal message contracts. Complex nodes can themselves be modeled as groups of nodes, giving the system a hierarchical architecture. Functionality is localised and data flow is visible. For humans and agents alike, the result is a concise architectural context that reads like a circuit diagram

But to be clear: a vmblu blueprint is not documentation, it is the system.

vmblu is designed for developing with coding agents. The blueprint is machine-readable, and is an explicit and concise context for the agents. Build the architecture first in collaboration between the human and the agent, and let the agent then write code against an agreed design. Not a trail of half-baked and ambiguous prompts, but a real architecture is the basis of a maintainable system.

At runtime, a vmblu application can be integrated into agent workflows in a consistent and standardized way. Node interfaces can be exposed as tools or events, enabling fine-grained natural-language interaction on top of the same architecture.

The message-based, asynchronous design makes these integrations straightforward, robust and verifiable.

You do not want to read or validate every line of code an agent has written, but you do want to understand what has been built. And you don't want technical debt to build up in your system.

vmblu makes systems inspectable. The code written by the agent is profiled and interfaces are checked against contracts. You can see what each node does, how data moves through the system and inspect when necessary. Behavior is explicit and localized. Bugs have a smaller blast radius and are easier for both humans and agents to trace.

Agents generate node-level tests and reports, so you do not rely on reading code to understand correctness.

vmblu includes a runtime for message switching between nodes. That runtime also acts as an enforcement layer: it applies security policies and exposes controlled capabilities to agents.

Security: nodes can be restricted in what they are allowed to access (network, filesystem, processes, etc.). If a node exceeds its permissions, the action is blocked and logged. Trust cannot rely on static analysis alone; vmblu enforces policy while the system runs.

Capabilities: external agents can interface with a vmblu application at a fine-grained level using mcp or other protocols. But capabilities allow an agent to do what it has to, ...and nothing more.

vmblu is released under the Apache-2.0 license—permissive and business-friendly.

Use it in commercial products, modify it freely, and contribute back to a growing ecosystem.