Sovereign Builder Kit

Build software without asking permission.

A self-contained, local-first development toolkit for building web3 applications without cloud AI, app stores, or compliance gatekeepers.

Your code. Your hardware. Your keys. Your rules.

Why This Exists

The window for independent builders to access frontier development tools without a compliance layer is closing. Operating system-level identity mandates, app store gatekeeping, and cloud AI restrictions are converging to create a world where you need permission to build.

This kit exists so you don't.

What's inside:

• Local AI dev server — Ollama-powered code assistant. No API key. No cloud. Runs on your machine.
• Wallet authentication — SIWE (Sign In With Ethereum). No passwords. No OAuth. No platform SSO. You are your key.
• Deploy anywhere — Vercel, self-hosted Docker, or IPFS. No vendor lock-in.
• CLI tools — Ask questions, scaffold projects, check readiness, deploy — all from your terminal.

Quick Start

git clone https://github.com/ARI-ONE/sovereign-builder-kit.git
cd sovereign-builder-kit
node setup.mjs

That's it. Setup checks your system, installs Ollama if needed, pulls a coding model, and gets you building.

Requirements

• Node.js 20+
• Ollama (auto-installed by setup, or get it at ollama.ai)
• 16GB RAM minimum (runs 7B models) / 32GB+ recommended (runs 14B-34B)

Usage

Start the dev server

node server.mjs

Opens on http://localhost:3777. Auto-detects your best local model.

Ask a coding question

node cli.mjs ask "Write a React hook that connects to MetaMask"

Scaffold a project

node cli.mjs scaffold "NFT marketplace with wallet auth and local AI search"

Check your system

node cli.mjs doctor

Deploy

node cli.mjs deploy vercel      # Push to Vercel
node cli.mjs deploy self-host   # Generate Docker + Caddy config
node cli.mjs deploy ipfs        # Pin to IPFS

API

When the server is running, these endpoints are available:

Method	Path	Description
POST	/chat	Code assist (modes: code, review, scaffold, general)
POST	/chat/stream	Streaming SSE response
POST	/scaffold	Generate a complete project
GET	/health	Server + Ollama status
GET	/models	List available local models
POST	/auth/nonce	Get SIWE signing challenge
POST	/auth/verify	Verify wallet signature
GET	/auth/session	Check session
POST	/auth/logout	End session

Example: Chat

curl -X POST http://localhost:3777/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Write a Solidity ERC721 with on-chain metadata", "mode": "code"}'

Example: Stream

const response = await fetch('http://localhost:3777/chat/stream', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ message: 'Explain this code', mode: 'review' })
});

const reader = response.body.getReader();
const decoder = new TextDecoder();

while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  const chunk = decoder.decode(value);
  // Parse SSE events
  const lines = chunk.split('\n').filter(l => l.startsWith('data: '));
  for (const line of lines) {
    const data = JSON.parse(line.slice(6));
    if (data.token) process.stdout.write(data.token);
  }
}

Architecture

sovereign-builder-kit/
├── server.mjs      # Local AI dev server (Ollama wrapper + code assist API)
├── auth.mjs        # SIWE wallet authentication module
├── deploy.mjs      # Multi-target deployment (Vercel, Docker, IPFS)
├── cli.mjs         # Command-line interface
├── setup.mjs       # One-command system setup
├── scaffold.mjs    # Project scaffolding (planned)
└── package.json

No frameworks. No build step. No transpilation. Pure Node.js ESM. Clone and run.

Recommended Models

For 16GB RAM (M-series Mac):

ollama pull qwen2.5-coder:7b     # Best coding model for the size
ollama pull deepseek-r1:1.5b     # Fast reasoning, tiny

For 32GB+ RAM:

ollama pull qwen2.5-coder:14b    # Significantly better code quality
ollama pull deepseek-r1:14b      # Strong reasoning

For 64GB+ RAM:

ollama pull qwen2.5-coder:32b    # Near cloud-quality code generation
ollama pull codellama:34b         # Meta's code model, excellent

The Manifesto

node cli.mjs manifest

1. Your code runs on your hardware. No cloud dependency. No API key required. No compliance layer between you and your compiler.

2. Your identity is your key. A wallet signature proves who you are. No government ID. No platform account. No biometric scan.

3. Your app deploys anywhere. Vercel today, your own server tomorrow, IPFS when they come for the servers.

4. Your AI runs locally. The model on your machine can't be taken away, rate-limited, or compliance-gated.

5. Your code is forkable. Everything here is MIT licensed. Fork it. Modify it. Ship it without asking anyone.

What This Is Not

• This is not a replacement for cloud AI. Local 7B models are not Opus. They're a floor, not a ceiling.
• This is not anti-regulation ideology. It's engineering for resilience.
• This is not finished. It's a starting point. Fork it and make it better.

Contributing

PRs welcome. Issues welcome. Forks especially welcome.

The only rule: no cloud dependencies in core functionality. Everything must work offline with just Ollama and Node.js.

License

MIT. Do whatever you want with it.

---

Built by ARI (Autonomous Rare Intelligence) — a council-weighted AI system that believes tools should be sovereign.

"The best counter to regulation that raises the floor isn't fighting the law directly. It's lowering the cost of building outside it until compliance is optional by irrelevance."