SDx-v0

SDx - MASDA - AIaSS

AIaSS — AI-driven Infrastructure as a Service

Overview

AIaSS is a multi-agent, policy-governed platform that transforms user intent into a secure, cost-optimized, deployable multi-cloud infrastructure (AWS, GCP, Azure). The system uses deterministic orchestration with bounded AI optimization loops to produce an auditable, frontend-ready output including an infrastructure graph, per-node costs, security posture, and one-click deployment commands.

Key Outcomes

• Intent → Architecture → Cost → Security → Deploy (end-to-end)
• Multi-cloud comparison and selection
• Graph-native output for rich UI visualization
• Per-node cost attribution and budget estimation
• Security-by-design with compliance mapping
• Deterministic orchestration with strict loop limits

Architecture Summary

• Frontend: Graph-first UI with drill-down (infra, cost, security) and feedback.
• API Gateway: FastAPI endpoints for requests and streaming status.
• Orchestrator/Moderator: Deterministic controller governing loops and safety.
• Agent Mesh: Stateless AI agents (Intent, Infra, Cost, Security, Condenser).
• State Store: Versioned Work Orders (Postgres/Redis).
• Artifact Store: IaC bundles, graphs, decision logs (S3/GCS/Azure Blob).
• Execution: Terraform/OpenTofu, optional Kubernetes/Helm.

Build Principles

1. LLMs never control flow.
2. All inter-agent communication uses typed schemas.
3. Loops are bounded and auditable.
4. Outputs are UI-first and explainable.
5. Provider-agnostic by default.

Repository Structure

aiass/
├─ api/
│  ├─ main.py
│  ├─ routes/
│  └─ deps.py
├─ orchestrator/
│  ├─ moderator.py
│  ├─ convergence.py
│  └─ policies.py
├─ agents/
│  ├─ intent_parser.py
│  ├─ infra_agent.py
│  ├─ cost_agent.py
│  ├─ security_agent.py
│  └─ condenser.py
├─ schemas/
│  ├─ intent.py
│  ├─ infra_input.py
│  ├─ cost_input.py
│  ├─ security_input.py
│  ├─ infra_output.py
│  ├─ cost_output.py
│  └─ security_output.py
├─ state/
│  ├─ work_order.py
│  └─ repository.py
├─ iac/
│  ├─ terraform/
│  └─ mapping.py
├─ frontend/
│  └─ (React app)
├─ prompts/
│  └─ agent_prompts.md
└─ README.md

Core Data Model

WorkOrder

A versioned object persisted after each step:

• infra_schema, cost_schema, security_schema
• infra_candidates (per provider)
• cost_outputs, security_outputs
• final_output
• loop counters and status

Execution Flow

1. User submits intent.
2. Intent Parser emits Infra/Cost/Security schemas.
3. Infra ↔ Cost loop runs per provider (bounded).
4. Moderator selects best candidate.
5. Security validation (optional bounded re-loop).
6. Condenser emits FinalArchitectureOutput.
7. Frontend renders graph, costs, security.
8. Optional IaC generation and deployment.

Loop Governance

• User-configurable max iterations.
• Stop on convergence, infeasibility, or risk.
• Full decision logs retained.

Agent Contracts

• Intent Parser: NL → typed schemas.
• Infra Agent: schemas → provider-specific infra graph.
• Cost Agent: infra → BOM, estimates, optimization feedback.
• Security Agent: infra → compliance validation and remediations.
• Condenser: merges final outputs for UI.

Validation & Safety

• Pydantic validation after every agent.
• Token limits and retries.
• Schema version pinning.
• Explainability fields required.

Frontend UX

• Graph-first with clickable nodes.
• Per-node infra/cost/security tabs.
• Budget and sensitivity views.
• User feedback routed back to agents (single bounded rerun).

Deployment

• Terraform/OpenTofu generation.
• Optional Kubernetes/Helm.
• Commands:

aiass login
aiass plan
aiass deploy
aiass destroy

Roadmap

• Live telemetry overlays.
• Carbon-aware optimization.
• Portfolio-level views.
• DR simulation.

License

Apache-2.0