Hierarchical Architecture
The system implements a Sequential Processing model. This ensures that every operation follows a logical, expert-vetted sequence from intent interpretation to the final response.The Collaborative Loop
This synergy is achieved through a specialized delegation pattern that ensures the “best expert” is always handling the task.1. Strategic Coordination
The Centurion Coordinator acts as the strategic architect. It analyzes the user’s request and builds a roadmap. If the roadmap involves different disciplines (e.g., investigating an IP and generating a report), it delegates tasks to the sub-agents.2. Expert Delegation (Fan-out/Fan-in)
- Fan-out: Centurion can trigger multiple sub-agents in parallel when tasks are independent.
- Expert Consult: An agent can “consult” another during its execution. For example, Scribe might query Perseus to gather inventory data for a report.
- Fan-in: Results from all specialized agents are aggregated back to the root for a unified analysis.
Shared Context & Memory
The secret to this synergy is Shared State. Unlike traditional systems where information is siloed, Myrmex agents share a “collaborative mental space.”Shared State
Unified Knowledge. Every finding from a sub-agent is added to the shared state, making it immediately available to subsequent agents in the execution chain.
Context Topology
Relationship Awareness. The Memory System allows agents to understand relationships (e.g., that a specific file belongs to the server currently being isolated).
Benefits of Hierarchical Synergy
- Reliability: By separating “Thinking” from “Doing” and “Formatting,” we ensure that actions are deliberate and responses are precise.
- Operational Efficiency: Agents only focus on what they are best at. Perseus handles the terminal, while Scribe handles the prose.
- Deep Specialization: Each sub-agent is trained with vendor-specific documentation and tools, providing an “Expert-level” performance in every domain.
This hierarchical approach allows Myrmex to tackle problems that are too complex for a single agent, delivering a cohesive experience that feels like a single, unified intelligence.