Homo Deus
Technical Standard

The autonomous AI substrate powering the entire AAMOS platform.

Authors: Erik Svensson · Johan Berglund · Wavult Group · Canonical as of 2026-05-17

1. Overview

Homo Deus is not a product. It is the substrate — the foundational automation layer on which all AAMOS products are built, deployed, monitored, and improved. It operates continuously without human intervention and is the reason AAMOS can guarantee its SLAs, certifications, and velocity.

2. Definition

Homo Deus is defined as the totality of autonomous systems that:

• Build and deploy software without human initiation
• Monitor all platform components and detect degradation
• Repair failures before they reach customers
• Improve the platform by analyzing patterns and implementing optimizations
• Maintain audit trails for all automated actions (GECL)
• Operate within defined constitutional constraints

Homo Deus is the answer to: "What runs when no human is watching?"

3. Product Hierarchy

Homo Deus sits below all products. Every AAMOS product is powered by — and built on — the Homo Deus substrate.

4. Components

4.1 Agent Runtime

Five specialized AI agents form the Homo Deus runtime. Each has defined responsibilities, trust levels, and constitutional constraints.

4.2 GECL — Governance, Ethics, Compliance, Logging

Every automated action performed by Homo Deus is logged through the GECL audit chain. GECL provides:

• Immutability: Logs cannot be modified or deleted retroactively
• Attribution: Every action is attributed to a specific agent and principal
• Evidence grade: Logs meet the evidence standard for TÜV AI Act certification
• Risk scoring: Each action is scored by risk level (E0–E5 scale)
• Human override: Any action can be halted by a principal with sufficient authority

4.3 Build Orchestration

Homo Deus uses multi-region AWS Bedrock to run parallel build jobs at scale:

• Up to 50 parallel agents per sprint
• Agent workers: EC2 r6i.2xlarge instances with wavult-bedrock-invoke IAM role
• Models: Qwen3-Coder 480B (heavy), Claude Sonnet (standard), Groq Llama (fast)
• Compute is on-demand — workers are created and terminated per sprint
• All build output committed to Gitea with signed commits

4.4 Self-Healing Infrastructure

Homo Deus continuously monitors the platform and takes corrective action:

5. Governing Principles

5.1 Human Oversight First

Homo Deus operates autonomously but humans remain in control. Any action with irreversible consequences — terminating infrastructure, public deployments, financial transactions — requires explicit principal approval.

5.2 Self-Hosted Only

All Homo Deus components run on Wavult-owned infrastructure. No external SaaS dependency is permitted for core automation. External cloud = AWS only.

5.3 Minimal Footprint

Agent workers are ephemeral. Compute is created for a task and terminated when done. No persistent state is stored on agent instances — only in Gitea, RDS, or S3.

5.4 Evidence-Grade Everything

Every automated decision must be traceable. If Homo Deus cannot explain why it took an action via GECL audit trail, it should not take the action.

6. Hard Constraints

These constraints cannot be overridden by any agent:

• No agent may exfiltrate private data outside Wavult infrastructure
• No agent may modify IAM policies, SCPs, or security group rules without principal approval
• No agent may initiate financial transactions above €500 without dual principal approval
• No agent may delete production databases or S3 buckets
• No agent may disable GuardDuty, CloudTrail, or CloudWatch monitoring
• Agent workers must stop within 12 hours if no active sprint is confirmed

7. Ownership