The problem Hedge Agent solves

HedgePilot makes crypto downside protection easier for everyday ETH holders.
Buying an on-chain option normally requires users to understand strike prices, expiries, collateral tokens, liquidity and smart-contract approvals. HedgePilot simplifies this process by allowing users to describe the protection they want in plain English or enter a few simple preferences.

The application then:

• Retrieves the current ETH price.
• Converts the user’s protection goal into clear trade parameters.
• Finds a matching live ETH put order through the Thetanuts SDK.
• Explains the cost and protection in plain language.
• Checks the contract address, approval amount and transaction parameters.
• Executes a small real transaction on Base mainnet only after explicit confirmation.

HedgePilot also includes a Guardian Mode prototype. It evaluates predefined market scenarios, distinguishes serious risks from less relevant events and explains whether protection may be appropriate. A future version could connect this reasoning to verified live news, market and on-chain data.

For hackathon safety, execution is capped at a small 1 USDC fill while still proving that the complete AI-to-mainnet transaction pipeline works.

Challenges we ran into

Short-lived marketplace quotes
Thetanuts marketplace signatures have a much shorter validity period than the option itself. Our first implementation retrieved an order before running the AI explanation and safety assessment. By the time the user reached the confirmation screen, the live quote could already be expired.

• We fixed this by changing the order of operations: the slower AI processing now completes first, and HedgePilot retrieves and previews the live marketplace order immediately before displaying the confirmation screen. We also added expiry validation and clearer retry messages.

JavaScript BigInt serialization
The Thetanuts SDK returns some blockchain values as JavaScript BigInt values. Express cannot directly serialize these values to JSON, which caused the quote request to fail with a Do not know how to serialize a BigInt error.

• We solved this by converting blockchain integers into strings or safe display numbers before returning them to the frontend.

Finding compatible live orders
Available option orders used different collateral assets and expiry periods. Some longer-duration orders required aBasUSDC, which would create an additional onboarding step for users holding native USDC.

• We filtered the live order book for ETH put orders using native Base USDC and selected the closest compatible strike and expiry. We chose the available 1/2/3-day markets for a simpler and more reliable demonstration.

RPC and network reliability
Calls to the Base RPC occasionally failed because of temporary network or TLS connection resets.

• We added retry handling and a fallback read-only Base RPC endpoint so temporary provider failures would not immediately break the quote flow.

Safe mainnet execution
The hackathon requires a real mainnet transaction, but we did not want the prototype to expose unnecessary funds.

• We therefore used a dedicated demo wallet, limited execution to a fixed 1 USDC fill, approved only the exact required amount and required explicit user confirmation before submitting a transaction.