Control Plane (UX)

The AEGIS Control Plane (aegis-control-plane) is a Next.js-based web application providing the primary operational and developer interface for the system. It abstracts the raw declarative YAML manifests and gRPC APIs into a visual, intent-driven experience.

Core Components

The Control Plane is divided into three major functional areas designed to support the complete agent lifecycle, from conception to observation.

1. The Architect

The Architect is the entry point for creating new agent workloads. Instead of writing Kubernetes-style AgentManifest YAML manually, users provide a natural language Intent (e.g., "Create an agent that monitors the staging database for slow queries and emails a report daily").

The Architect parses this intent and automatically generates a complete manifest, configuring:

• The appropriate standard or custom runtime.
• A structurally sound SecurityPolicy (including precise network and filesystem allowlists).
• The necessary ResourceLimits and role assignments.

2. The Synapse

The Synapse is the real-time execution visualizer. Once an agent execution begins, The Synapse connects to the orchestrator's event stream to render the 100monkeys iterative refinement loop (Execute → Evaluate → Refine) live.

Transparent Iteration is the foundational UX principle of The Synapse. Unlike traditional chat interfaces that hide the intermediate reasoning steps and only present the final result, The Synapse exposes:

• Every tool invocation and its resulting output (e.g., cmd.run, fs.read).
• Validation scores and judge feedback for each iteration.
• The precise code diffs or strategy changes applied during the Refine phase.
• Policy violations blocked by the SMCP Gateway.

This transparency allows operators to understand how an agent arrived at a solution and where it struggled.

3. The Cortex Explorer

The Cortex Explorer is a memory browser that visualizes the system's learned intelligence. While the orchestrator handles the mechanics of communicating with Cortex (or running in memoryless mode), the Explorer provides the human interface to:

• Browse learned Patterns and the specific error signatures that trigger them.
• Track the evolution of higher-level Skills.
• View confidence metrics (SuccessScore) and memory decay status.

Integration with the Orchestrator

The Control Plane communicates with the AEGIS Orchestrator exclusively via the public gRPC and HTTP APIs. It does not have direct access to the orchestrator's internal state, databases, or the underlying container runtimes.

// Conceptual API surface consumed by the Control Plane
interface ControlPlaneAPI {
  parseIntent(intent: UserIntent): Promise<ParsedIntent>;
  generateManifest(intent: ParsedIntent): Promise<AgentManifest>;
  deployAgent(manifest: AgentManifest): Promise<AgentId>;
  streamExecution(executionId: ExecutionId): AsyncIterable<ExecutionEvent>;
}

Authentication between the Control Plane and the Orchestrator is handled via OIDC tokens issued by the central IAM system (Keycloak), ensuring that operators only see and control the agents within their authorized tenant boundary.