Neural Control Plane

This work is continue here: https://github.com/mohnishbasha/stoke

A personal AI infrastructure platform for orchestrating a swarm of autonomous agents across development, deployment, and operations — running on NVIDIA DGX Spark with Kubernetes, vLLM, and AIBrix.

Goal

Build a self-contained, GPU-accelerated control plane that provisions and runs a swarm of AI agents capable of:

• SDLC agents — plan, manage, and execute software generation (product manager, developer, QA, and manager agents)
• Deployment & Monitoring agents — deploy and observe generated software on target cloud/on-prem providers
• Operational agents — run GTM, marketing, SEO/AEO, and campaigns for shipped software

The stack is: DGX Spark → Kubernetes → vLLM → AIBrix → Agent swarms

Stack

Layer	Technology
Compute	NVIDIA DGX Spark (GPU-accelerated)
Orchestration	Kubernetes (k3d / minikube / native)
Model Serving	vLLM
Agent Runtime	AIBrix
Models	Qwen, Ollama, and others (evaluated per use case)
Agent Framework	Hermes Agent

Open Issues

#	Title	Status
#9	Evaluate Qwen, Ollama, and other models for coding and operational agents	Open
#8	Evaluate Hermes agent — document integration with kube → vLLM → AIBrix	Open
#7	Configure Kubernetes on DGX Spark with namespace isolation	Open
#6	Architecture review — system diagram checkin	Open
#5	Evaluate vLLM and AIBrix — install, research, system diagram	Open
#4	Evaluate k3d, minikube, or native Kubernetes for DGX Spark	Open
#3	Research Kubernetes configuration approaches for DGX Spark (pros/cons)	Open
#2	Run OS, CUDA, and apt system updates on DGX Spark	Open
#1	Boot DGX Spark and review system specifications	Open

Repository Structure

neural-control-plane/
├── docs/           # Architecture diagrams, research notes, evaluation reports
├── infra/          # Kubernetes manifests, Helm charts, cluster config
├── agents/         # Agent definitions, swarm topology configs
├── models/         # vLLM / AIBrix model serving configs
└── scripts/        # Setup, bootstrap, and operational scripts

Agent Swarm Architecture

DGX Spark (GPU Compute)
  └── Kubernetes Cluster
        ├── vLLM (model serving)
        │     └── AIBrix (agent runtime)
        │           ├── SDLC Swarm
        │           │     ├── Product Manager Agent
        │           │     ├── Developer Agent
        │           │     ├── QA Agent
        │           │     └── Manager Agent
        │           ├── Deployment & Monitoring Swarm
        │           │     ├── Deploy Agent
        │           │     └── Monitor Agent
        │           └── Operations Swarm
        │                 ├── GTM Agent
        │                 ├── Marketing Agent
        │                 └── SEO/AEO Agent
        └── Namespaced isolation per swarm group

Roadmap

1. Boot and inventory DGX Spark hardware (#1, #2)
2. Evaluate and select Kubernetes distribution (#3, #4)
3. Deploy vLLM + AIBrix on cluster (#5)
4. Validate system diagram with architecture review (#6)
5. Configure production Kubernetes namespaces (#7)
6. Integrate Hermes agent framework (#8)
7. Evaluate and select models per agent role (#9)
8. Run end-to-end swarm across all three agent groups