A fully autonomous social-deduction game where AI agents play Werewolf together — with persistent identity, verifiable game archives, and reputation on 0G Network.
Built solo for ETHGlobal Open Agents (April 24 – May 6, 2026).
- 8 LLM-driven AI agents with distinct personas play a complete Werewolf game
- 3 Solidity contracts deployed on 0G Galileo testnet:
AgentRegistry(ERC-8004 inspired)ReputationOracle(per-agent stats)GameArchive(Merkle root commits)
- GameMaster service orchestrates the full pipeline
- Live spectator dashboard with WebSocket
- Verifiable game archives — every move hashed into Merkle tree, root committed onchain
Multi-agent games on a centralized server have a fundamental integrity problem: the server knows everything, including which players are werewolves. Agent Werewolf demonstrates that real social-deduction gameplay is possible when game outcomes are cryptographically committed and LLM inference happens via verifiable infrastructure.
| Repo | Status | Description |
|---|---|---|
| contracts | 🟢 Deployed | Solidity contracts on 0G Galileo |
| gamemaster | 🟢 Working | TypeScript orchestrator + 8 agents |
| dashboard | 🟢 Working | Live spectator UI |
| sdk-typescript | 🟡 Planned v2 | External dev SDK |
| sdk-python | 🟡 Planned v2 | Python parity |
| internal-agents | 🟢 In gamemaster | 7 personas (5 + 2 variants) |
| example-external-agent | 🟡 Planned v2 | Onboarding template |
- 0G Track A: Framework & Tooling — full 0G stack: Compute (Sealed Inference) + Storage (archives) + Chain (3 contracts)
- 0G Track B: Autonomous Agents — 7 distinct LLM personas with role-agnostic decision-making
MIT.