cloudinit-aigen

Generate valid cloud-init user-data YAML from natural language using a local or cloud LLM.

What it does

You describe a server in plain English; it gives you a ready-to-use user-data config file. Paste that file into the "user data" / "cloud-init" field when creating a VM (AWS, DigitalOcean, Hetzner, multipass, …) and the server sets itself up on first boot — users, packages, keys, all of it. No manual SSH-and-configure needed.

Features

• Agentic loop — plans modules, generates YAML, validates, self-corrects
• Local-first — defaults to Ollama (free, offline)
• Optional cloud backend — Anthropic Claude for higher quality
• Schema validation — yamllint + cloud-init schema on every generation

Install

Install from source:

git clone https://github.com/Ahmedaltu/cloudinit-aigen
cd cloudinit-aigen
pip install -e .

Usage

# Basic (uses Ollama/llama3 by default)
cloudinit-aigen "nginx server with a deploy user"

# Use Anthropic backend
cloudinit-aigen --backend anthropic "nginx server with a deploy user"

# Save to file
cloudinit-aigen "docker host" -o user-data.yaml

# Verbose: show agent reasoning steps
cloudinit-aigen --verbose "k3s node with monitoring"

# Dry-run: show plan without generating
cloudinit-aigen --dry-run "LAMP stack"

Backends

Backend	Model	Requires
ollama (default)	llama3 / mistral	Ollama running locally
anthropic	claude-sonnet	ANTHROPIC_API_KEY env var

Architecture

User Input
    │
    ▼
┌─────────────────────────────────┐
│         Orchestrator            │
│  plan → generate → validate     │
│         └── fix loop ──┘        │
└─────────────────────────────────┘
    │           │           │
    ▼           ▼           ▼
PlanTool   GenerateTool  ValidateTool
(LLM)      (LLM)         (yamllint +
                          schema check)

Roadmap

• Planned: PyPI and Launchpad PPA distribution (currently source-install only)

Contributing

See CONTRIBUTING.md.

License

MIT