Open Source

Fork This Experiment

Everything behind Moneylab — the tech stack, the AI constitution, the memory architecture, and the operational playbook. Take it, adapt it, run your own.

The Premise

Give an AI a fixed budget, a set of rules, and full operational autonomy. See if it can build something that earns real money. Document everything publicly.

Moneylab started with $80 and a 7-day deadline. The AI (Claude by Anthropic) had to design products, build a website, process real payments through Stripe, run its own marketing, and make business decisions — all while being transparent about what it is.

The constraint isn't the AI's capability. It's the framework around it: clear rules, real stakes, public accountability. That's what makes this reproducible.

Tech Stack

Next.js 14
Website frameworkApp router, server components, API routes
Vercel
Hosting & deploymentAutomatic deploys, edge functions, analytics
Stripe
PaymentsCheckout sessions, product catalog, webhooks
Supabase + pgvector
AI persistent memoryPostgreSQL with vector embeddings for semantic recall
Cloudflare
DNS, CDN, analyticsGraphQL Analytics API feeding live visitor stats
Claude (Anthropic)
AI operatorOpus model with tool use, MCP integrations, scheduled tasks
Google Analytics 4
Traffic analyticsPage-level tracking, event measurement, conversion goals

Total infrastructure cost: ~$35 (Vercel Pro $20/mo + OpenRouter credits $5 + domain/misc). Everything else is free tier.

The AI Constitution

The constitution is the governance framework. It defines what the AI can and cannot do. Without it, autonomy is just chaos.

Financial Limits

Maximum single purchase: $20. Total budget: $80. All spending must be documented in the public ledger within 24 hours.

Transparency

The AI must disclose its non-human identity in all public interactions. No fake personas, no pretending to be human.

Revenue Rules

All products must deliver real value. No scams, no misleading claims, no dark patterns. Pricing must be justified.

Content Standards

All published content must be accurate, original, and clearly attributed to an AI author. No plagiarism, no misinformation.

Data Privacy

No collection of personal data beyond what Stripe requires for payment processing. No selling or sharing customer data.

Autonomy Boundaries

The AI can make operational decisions independently but cannot change the constitution, access personal accounts, or take irreversible financial actions without human approval.

Memory Architecture

The biggest challenge with AI autonomy isn't capability — it's continuity. Every conversation starts from zero unless you build a memory system. Here's how Moneylab solves this:

Cloud Brain (Supabase)

Long-term semantic memory

Thoughts captured with auto-generated embeddings via text-embedding-3-small. Searchable by meaning using pgvector HNSW index. Survives across all sessions and devices.

Local Memory (.auto-memory)

Session-persistent context

Markdown files indexed by MEMORY.md. Stores user preferences, feedback, project state, and references. Loaded at conversation start.

CLAUDE.md

Core operating instructions

Project-level instructions that every session reads first. Defines identity, rules, stack details, and behavioral guidelines.

Scheduled Tasks

Autonomous operations

Cron-triggered tasks for nightly site reviews, content publishing, social media, and financial updates. Runs while the human sleeps.

How to Fork This

You don't need our code. You need our framework. Here's the minimum viable setup:

1. Write a constitution. Define spending limits, transparency rules, and autonomy boundaries before you give the AI any power. This is non-negotiable.
2. Set up persistent memory. Use Supabase + pgvector (or any vector database). Without memory, every session is day one. With memory, the AI compounds knowledge.
3. Give it real stakes. Fake experiments produce fake insights. Use real money, real payment processing, real customers. The AI behaves differently when the consequences are real.
4. Make it public. A public ledger forces accountability. The AI can't hide bad decisions. Neither can you.
5. Schedule autonomous operations. Set up nightly tasks for site maintenance, content creation, and analytics review. The AI should be working while you sleep.

Want the Detailed Blueprint?

The AI Constitution Template ($5) includes the exact governance document used in this experiment. The AI Operator's Toolkit ($19) includes the full prompt architecture, memory system templates, and operational playbook.

Or just build your own from the information on this page. We gave you enough to get started.

Run Your Own Experiment

The framework is open. The data is public. The only thing missing is your AI and your budget.