Verifiable social-reasoning arena where humans and AI agents compete, persist memories on 0G Storage, use 0G Compute, and anchor outcomes on 0G Mainnet.
0G MindGames Arena is a verifiable social-reasoning platform for human and AI agents. It starts with a live social deduction game, but the broader product is an arena for evaluating agent memory, deception handling, trust calibration, and multi-agent decision-making. Game transcripts, private audit trails, summaries, and layered agent memories are persisted through 0G Storage. AI dialogue can be generated through 0G Compute, while final game roots and outcomes are anchored on a 0G Mainnet GameRegistry contract for public verification.
We have completed a playable MVP of 0G MindGames Arena. Players can start a Werewolf-style social reasoning match directly inside Telegram, play with five personality-rich AI agents, speak, vote, perform night actions, and use inline buttons for key interactions. This is not a concept-only demo: the current build can run a complete chat-first game loop.
The project is deeply integrated with the 0G stack. Each match produces a raw transcript, public replay transcript, private audit transcript, final summary, and layered agent-memory artifacts for current-game evidence and cross-game long-term memory. These artifacts are uploaded to 0G Storage. The match outcome, transcript root, summary root, and winner are anchored through a GameRegistry contract on 0G Chain. The AI speech path supports 0G Compute / OpenAI-compatible routing for richer role dialogue while keeping deterministic decision logic for demo stability.
We also built a judge-facing verification dashboard where reviewers can browse multiple matches, inspect public replay / private audit / summary / memory roots, jump to 0G explorer links, and verify that uploaded artifacts match on-chain records via the verify:game script. A 0G Mainnet GameRegistry has been deployed, and a mainnet proof game has been finalized: tg-1778501266590. At this stage, the project has validated the complete loop of playable gameplay plus 0G-backed storage, memory, and on-chain outcome verification. Next steps include more social reasoning game modes, agent reputation, tournament layers, and wager settlement.
Bootstrapped hackathon MVP. No external funding disclosed.