Ouroboros — Describe a Game, Get a Deployed Pyth dApp (AI AGENT)

Pyth Community Hackathon
1

Project Name: Ouroboros

Team: pstar

Submitted: April 1, 2026

Answer Capsule

Ouroboros is an autonomous AI coding agent that builds Pyth-powered mini-games from plain English descriptions. It uses Pyth Price Feeds via the Hermes API for real-time asset pricing in prediction and trading games, and Pyth Entropy for verifiable on-chain randomness in dice, card, and loot-drop games — deploying smart contracts to Base Sepolia and frontends to Netlify, all through a chat interface.

What It Does

Most developers face a steep learning curve when building on-chain games that use real-world data or verifiable randomness. Ouroboros eliminates that barrier entirely. A user describes a game idea in plain English — “build me a coin flip game using ETH price” — and an AI agent autonomously writes the smart contracts, builds the frontend, integrates Pyth feeds, and deploys everything live. No Solidity knowledge required. No React experience needed. Just an idea and a chat window.

The agent operates through a ReAct (Reasoning + Acting) loop: it thinks about what to do, executes tools (writes files, runs terminal commands, fetches Pyth data, deploys contracts), observes the results, and iterates — all streamed to the user in real time.

Pyth Features Used

  • Price Feeds (on-chain/off-chain via Hermes API)

  • Entropy (on-chain verifiable randomness)

  • Both

Price Feeds: The agent uses Pyth Hermes API (hermes.pyth.network) to fetch real-time prices for any supported asset. When building games, it integrates price data directly into frontends and smart contracts — enabling prediction markets, trading simulators, price-reactive gameplay mechanics, and more. The agent has built-in tools for price fetching, feed discovery, candlestick (OHLC) data, and historical price lookups.

Entropy: For games requiring randomness (dice rolls, card draws, loot drops, coin flips), the agent deploys Solidity contracts to Base Sepolia that integrate with the Pyth Entropy contract at 0x4821932D0CDd71225A6d914706A621e0389D7061. Contracts implement the IEntropyConsumer interface for callback-based verifiable randomness — ensuring fair, tamper-proof outcomes on-chain.

Links

Screenshots/Media

coinflip
coinflip2492×1320 278 KB

Tech Stack

| Layer | Technology |

|-------|-----------|

| Backend Framework | Python, FastAPI, Uvicorn |

| Frontend Framework | Nuxt 3, Vue 3, TypeScript |

| Styling | Tailwind CSS, Nuxt UI 3 |

| Database | MongoDB Atlas (async via Motor) |

| LLM Providers | Anthropic Claude, OpenAI GPT |

| Blockchain | Solidity, Foundry, Base Sepolia |

| Pyth Integration | Hermes API (prices), Entropy (randomness), MCP Server (feed discovery) |

| Deployment | Netlify (frontend + generated games), VPS with nginx + systemd (backend) |

| Charts | ApexCharts |

How It Works

  1. Create a project — Pick a template: Entropy Game (on-chain randomness), Price Game (live Pyth feeds), or Custom Game

  2. Describe your game — Chat with the agent: “Build a dice game where the payout multiplier is based on the current ETH/USD price”

  3. Watch it build — The agent writes smart contracts, builds a React frontend, installs dependencies, integrates Pyth, and deploys — all streamed live

  4. Play your game — Click the live preview link to play the deployed game immediately

  5. Iterate — Keep chatting to add features, tweak mechanics, or redeploy

The agent has 10 specialized tools:

  • File operations: read, write, patch, search

  • Terminal: execute shell commands with streaming output

  • Pyth tools: price fetching, feed search, candlestick data, historical prices, contract deployment

Each tool call is visible in the activity log so users can follow along and learn how their game is being built.

How to Use It (For Judges)

  1. Enter your own Anthropic API key (Claude) in the header bar — select Claude Sonnet
  2. Create a new project — pick any template (Entropy Game, Price Game, or Custom)
  3. Describe your game idea in the chat, e.g. “Build a coin flip game where the payout is based on the current ETH price”
  4. Wait for the agent to finish — it may appear stuck at times (especially during installs or deploys) but it will recover and continue. Give it a couple minutes before retrying.
  5. Once done, click the live preview link to play your deployed game.
  6. Don’t be shy to tell the agent that the live preview did not work correctly and continue fixing deployment

Tip: The agent works best with Claude Sonnet. If it seems unresponsive for 30+ seconds, the stuck detection will kick in — you can either wait for it to self-recover or hit “Force Stop” and send a follow-up message like “continue where you left off.”

Content Contributions

Licensing

This project is licensed under Apache 2.0.

Eligibility Confirmation

  • I am 18 years of age or older

  • I am not located in an OFAC-sanctioned territory

  • This is original work created during the hackathon period

  • I agree to the Terms & Conditions