Architecture

Omnipotent is modeled after operating system architecture. A Kernel manages agents, a Signal Bus handles coordination, and tiered memory ensures nothing is forgotten.

System Overview

Memory Tiers

Hot Memory (The Focus)

• Mechanism: Local session-mapped storage
• Purpose: Immediate task context. "What am I doing right now?"
• Rationale: Keeping immediate context local minimizes latency and ensures that sensitive, in-progress work is never exposed to external databases.

Cold Memory (The Grimoire)

• Mechanism: Vector-embedded relational storage
• Purpose: Long-term history and semantic search. "How did we solve this bug last month?"
• Rationale: By using semantic similarity search, agents can recall patterns across vast historical datasets without needing the user to provide explicit file pointers.

Federated Memory (The Library)

• Mechanism: Secure cross-realm artifact exchange
• Purpose: Team wisdom. "Has anyone else in the Alliance solved this?"
• Rationale: Allows specialized agents to share high-level strategic plans and patterns while maintaining strict cryptographic isolation between sovereign domains.

The Kernel

The Kernel manages the agent lifecycle and enforces the system constitution. It is responsible for:

• Booting and registering the agent collective
• Enforcing signal integrity and authentication
• Managing the secure visual dashboard interface
• Coordinating secure container boundaries

System Core Structure

The system core is organized into specialized layers: a Service Layer for coordination, a Visual Layer for the human-in-the-loop interface, and a Native Layer written in memory-safe systems languages for performance-critical security operations.

The Signal Bus

All inter-agent communication flows through a Redis pub/sub bus. No agent talks to another directly. This enforces loose coupling and makes the system observable.

• Channel pattern: omnipotent:{agent}:{event}
• Example: omnipotent:shadow-route:task-dispatched
• Monitoring: The dashboard subscribes to all channels for real-time visibility

The Swarm Blackboard

Agent coordination uses a blackboard pattern via .realm/telemetry/swarm_state.json. Agents read and write to this shared file to track:

• Current task assignments and status
• Agent health and error counts
• Blockers and dependencies
• Execution phase (planning, executing, reviewing, archiving)

Docker Network Topology

Related

• Agents — All 12 roles and their responsibilities
• Protocols — Mnemosyne, Airlock, Swarm, Diplomat
• Federation — Sovereign vs Alliance domain architecture