Architecture Overview

The VOX architecture is built upon a decoupled, event-driven model that prioritizes security, observability, and deterministic execution. It moves away from monolithic scripts toward a specialized, audited hierarchy.

1. Agents: The Identity Layer

In the VOX ecosystem, an Agent is an independent execution unit defined by a unique identity and personality.

• Autonomy: Agents are not merely passive functions; they are stateful entities that manage their own lifecycle and subscriptions within the orchestrator.
• Orchestration: Each Agent acts as a container for specialized behaviors, ensuring that logical concerns (e.g., Security Operations vs. Personal Assistance) remain isolated.

2. Roles: The Behavioral Layer

A Role is a modular behavioral unit that defines an Agent's expertise. In VOX, Roles are subject to mandatory static analysis before instantiation.

• Contextual Intelligence: Roles govern the system prompts and internal logic, ensuring strict adherence to executive protocols.
• Event-Driven Synapsis: Roles interact via an internal event bus, allowing for complex multi-agent cooperation without direct coupling.
• Hardened Execution: By keeping logic within Roles and I/O within Capabilities, the behavioral layer remains clean and auditable.

3. Capabilities: The Resource Layer

The Capability is the atomic unit of system permission, following the Object-Capability (OCAP) Model.

• Zero Trust: An Agent possesses no inherent system power. Access to restricted resources—such as the filesystem, network, or LLM inference engines—is explicitly granted via secure Capability injection.
• Managed I/O: Capabilities act as the only "Gateways" to the host system. This ensures that all I/O is asynchronous, logged, and sandboxed, preventing unauthorized system access or event-loop blocking.