Crucible is a trustless proving ground for DeFi agents. It forces autonomous AI to survive simulated extreme market conditions before they can trade real capital or mint on-chain attestations.
Crucible is a trustless proving ground and testing gauntlet designed to solve the biggest problem in Web3 AI: Agentic Risk.
As developers build autonomous AI agents to manage DeFi portfolios, the risk of hallucination, slippage exploitation, and parameter violation grows exponentially. Crucible allows protocol engineers to design "Custom Arenas"—simulated, deterministic market scenarios (e.g., Flash Crashes, Oracle Attacks, Depeg Events). Developers then stake testnet ETH and subject their agents to these brutal market simulations.
Powered by Azure OpenAI (gpt-5.4) as the execution brain, the agent's real-time decisions are evaluated against strict on-chain risk bounds. If the agent survives, it receives a verifiable Attestation (SBT) proving its reliability. If it violates risk parameters (e.g. accepts 5% slippage during a crash), the Crucible Guard instantly blocks the action and slashes the agent's staked ETH. Crucible ensures agents are battle-tested before they are allowed anywhere near real capital.
Everything was built entirely from scratch during this Hackathon!
Smart Contracts: Developed and deployed the core Crucible.sol protocol on Arbitrum Sepolia for registering agents, publishing custom arenas, and handling slashing/attestations.
AI Orchestrator Engine: Integrated Azure OpenAI to act as the cognitive engine, generating deterministic JSON trading decisions based on simulated market tick events.
Custom Arena Builder: Built a UI allowing users to dynamically prompt-engineer and generate their own extreme market environments.
Web3 Infrastructure: Integrated Wagmi, Viem, and RainbowKit for a seamless wallet connection and on-chain transaction execution.
Bootstrapped. We are currently evaluating early-stage / Pre-Seed opportunities to bring Crucible Protocol to mainnet and integrate decentralized Validator nodes (AVS) for the AI execution.