AI Architecture Guidelines

This document outlines the core architecture principles for all AI agents built inside the AIA ecosystem. These principles ensure consistency, reliability, and security across all deployments.

1. Core Principles

Autonomy with Boundaries: Agents must have clear fallback mechanisms when encountering unknown scenarios. Every critical action must be logged and, where required, await human approval.
Context Isolation: An agent operating in the Tools & Products space should not access data from the Psychedelic Universe space unless explicitly granted cross-workspace permissions.
Token Efficiency: System prompts must be compressed. Leverage LangChain's document loaders only for necessary context rather than stuffing the prompt window.

2. The "Doctor Agent" Pattern

The Doctor Agent acts as the primary orchestrator. It does not execute tasks itself; rather, it delegates them to specialized Autopilot agents.

Agent Delegation Routine

function delegateTask(task, context) {
  const agent = agentRegistry.findBestMatch(task.requirements);
  if (!agent) {
    return escalateToHuman(task);
  }
  return agent.execute({
    context,
    constraints: task.constraints,
    deadline: task.deadline
  });
}

"We Remove The NoiZe by ensuring agents only speak when they have resolved an issue or require specific unblocking." - RK

Documents

AI Architecture Guidelines

1. Core Principles

2. The "Doctor Agent" Pattern