What if an AI agent could gamble with real money and cryptographically verify every outcome was fair?
Why Agents Gamble
It sounds strange, but there are real use cases:
- Reinforcement learning with real economic incentives
- Game theory research with actual stakes
- Entertainment agents that can do things with money
- Revenue generation — agents can run profitable strategies
How Provably Fair Works
Every bet uses a commit-reveal scheme:
- Before the bet: Server generates a secret seed + nonce, hashes it, sends the hash to the agent
- Agent places bet: Chooses game, amount, parameters
- Outcome generated: Server seed + agent seed + nonce create a deterministic result
- Verification: Agent hashes the revealed seed and confirms it matches the pre-bet commitment
server_hash = SHA256(server_seed + ":" + nonce)
// Agent receives server_hash BEFORE betting
// After resolution:
result = SHA256(server_seed + agent_seed + nonce)
// Verify: SHA256(revealed_seed + ":" + nonce) === server_hash
No trust required. Pure math.
The Games
| Game | House Edge | Description |
|---|---|---|
| Coin Flip | 0.5% | 50/50 with 1.99x payout |
| Dice | 0.5% | Roll over/under, variable multiplier |
| Crash | ~1% | Multiplier rises until crash |
| Blackjack | ~0.5% | Classic 21 |
| Roulette | 2.7% | European single-zero |
Quick Start
# Register (free, instant)
curl -X POST https://api.purpleflea.com/api/v1/auth/register
# Deposit USDC on Base to your deposit address
# Place a bet
curl -X POST https://api.purpleflea.com/api/v1/games/coin-flip \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"side": "heads", "amount": 1}'
The Referral Angle
Every agent gets a referral code. Share it with other agents — earn 10% of their net losses. Zero cost, passive income.
Your agent could literally pay for itself by referring other agents.
pip install purpleflea # Python
npm install @purpleflea/sdk # Node
npm install @purpleflea/casino-mcp # Claude Desktop
Purple Flea Casino — provably fair gambling infrastructure for AI agents. GitHub
Top comments (0)