The genesis of the Agentic Farm. This repository contains the early AI infrastructure and multi-agent experiments that laid the foundation for everything that followed — built from September 2023, fourteen months before the market started talking about AI agents.
This is an archive. CloudMind represents the earliest experiments — the raw prototypes and infrastructure that evolved into Moulinsart (September 2025) and the production Ferme Agentique (2026).
| Date | Milestone |
|---|---|
| Sept 2023 | CloudMind begins — Kubernetes, vector databases, AI infrastructure |
| Q1 2024 | Project Nexus — dynamic 4-agent team composition with CrewAI |
| Oct 2024 | Exam Genius — multi-agent education platform |
| Sept 2025 | Moulinsart — first production agentic farm (tmux + email) |
The AI Architect — a system where a Master AI reads project mandates (PRD markdown files) and dynamically composes 4-agent teams:
- MasterAI reads the mandate, designs the team
- 4 agents execute sequentially with accumulated context
- Agent groups dissolve after mission completion
- Auto AI Box Generator: enter a use case name → generates YAML → auto-creates Python files (agents.py, tasks.py, workflow_function.py)
- CrewAI hierarchical process with GPT-4 as manager
User enters use case → YAML generated → Python files created →
→ Fetcher Agent (search) → Analyzer Agent (format) →
→ Proxy Agent (supervise) → Compiler Agent (compile) →
→ Markdown output saved
- NOVASPHERE: Agnostic AI framework supporting OpenAI, Anthropic, and HuggingFace backends
- AegisMind: Multi-agent workflow with Coordinator, Engineer, Debug, and Design agents
Full cloud-native AI deployment:
- GKE + Azure AKS orchestration
- JupyterLab pods with dynamic provisioning
- Streamlit apps deployed as K8s services
- Pinecone vector databases (1,500+ vectors, 21 dimensions)
- LiveKit speech-to-speech integration
- Grafana + InfluxDB monitoring dashboards
- Interactive deployment toolbox (
deploy_k8s.sh) with 7 operations
7+ hours per day dedicated to building AI systems from scratch:
- Docker orchestration
- Kubernetes pod management
- Vector database design
- Real-time communication (LiveKit)
- Multi-agent framework experimentation
| File | Purpose |
|---|---|
AI Equalizer.py |
Project Nexus — MasterAI + dynamic agent team composition |
defpromptcrew.py |
CrewAI integration + Auto AI Box Generator (46KB) |
defyaml.py |
YAML config generator for agent workflows |
deploy_k8s.sh |
Interactive Kubernetes deployment toolbox |
app.py |
Streamlit file uploader for K8s pods |
scripts/ |
K8s management scripts (deploy, status, cleanup) |
shell-automation/ |
Automated deployment verification |
- Python 3.9+
- Kubernetes CLI (
kubectl) - OpenAI API key
- Serper API key (for search tools)
- Docker
CloudMind proves that the agentic approach — AI agents composing teams dynamically, executing mandates, dissolving after completion — was being built here in Q1 2024. This is 14 months before multi-agent frameworks became a mainstream topic, and 18 months before OpenClaw's first commit.
The DNA of the Ferme Agentique is visible in every file: mandate-driven execution, agent specialization, hierarchical coordination, and automated workflow generation.
MIT License. See LICENSE for details.
Built by Mr D — Founder of Infinity Cloud, Switzerland.