Sentient

The Problem: Dead Capital in DeFi
Right now, over $50 billion in DeFi liquidity is completely dead capital. Static 150%+ collateral ratios trap this wealth because smart contracts are fundamentally blind to real-world risk and off-chain creditworthiness. This massive overcollateralization makes DeFi far too expensive for borrowers in emerging markets where capital efficiency is needed most.
The Solution: Sentient
Sentient is a next-generation decentralized finance protocol built on Base Sepolia. By leveraging the brand new Chainlink Runtime Environment (CRE v1.1.2), we use decentralized, off-chain AI agents to dynamically and safely lower collateral requirements to as low as 105%.
This isn't just a feature; it is a financial breakthrough. By reducing collateral limits by 45%, Sentient unlocks billions of dollars in dormant capital for the developing world, creating unprecedented capital efficiency.
Multi-Agent Architecture (Hackathon Tracks)
We engineered a highly modular architecture utilizing four independent CRE agents to execute on multiple hackathon bounties:

1. The AI Actuary Agent (Chainlink CRE Main Track) A CRON-triggered agent that reads a user's on-chain health factor, runs a macroeconomic volatility scan via Google Gemini 2.5 (monitoring emerging markets like NGN and TRY), and dynamically adjusts the smart contract's collateral ratio based on real-world stability.
2. The Privacy Shield (Privacy Track) To offer 105% ratios, we need to verify off-chain fiat reserves. This agent uses CRE's Confidential HTTP capability to ingest Web2 bank credentials off-chain. It checks fiat reserves but explicitly redacts the financial data in memory. It reaches consensus and only broadcasts the final 105% ratio integer on-chain, leaking zero sensitive data.
3. The Mini App Bridge (World ID Track) World Apps are natively restricted to World Chain, but Sentient's liquidity lives on Base Sepolia. To solve this, our React frontend acts as a World Mini App, generating a ZK proof via MiniKit, and POSTs it to our CRE Agent. The agent verifies the proof off-chain via the Worldcoin API, and acts as a cross-chain bridge, rewarding the verified human on Base Sepolia.
4. The Moltbook Publisher (Agents-Only Track) An autonomous CRON agent that synthesizes daily protocol risk metrics using Gemini 2.5 and automatically POSTs its own status updates and risk reports directly to the m/chainlink-official community on the Moltbook network using the CRE HTTPClient.
   The Tenderly War Room (Sponsor Track)
   Giving an AI power over DeFi liquidity is dangerous. Before our CRE agents are allowed to write their dynamic ratio adjustments to Base Mainnet, we route their transactions to a Tenderly Virtual TestNet shadow-fork. We are completely sandboxing our AI execution for enterprise-grade safety.
   Why It Matters
   Sentient bridges TradFi data, World ID Sybil resistance, and AI underwriting. orchestrated securely through Chainlink CRE. We aren't just building a smart contract; we are unlocking the next generation of capital-efficient DeFi for the developing world.

1. The Trigger Engine (Cron)
   At the very bottom of the file (initWorkflow and onCronTrigger), the agent uses Chainlink's CronCapability. Instead of waiting for a user to click a button, this agent wakes up automatically based on a schedule (e.g., once a day) to execute its logic.
2. Decentralized Execution (runInNodeMode)
   The core logic is wrapped inside runtime.runInNodeMode(..., consensusIdenticalAggregation). In a live Chainlink network, this forces multiple decentralized nodes to run the exact same code, hit the APIs, and reach an agreement (consensus) before proceeding.
3. Step One: AI Synthesis (Gemini API)
   Inside synthesizeAndPublish, the agent takes your protocol's stats (TVS, Health, Ratio) and uses the CRE HTTPClient to securely ping the Google Gemini 2.5 Flash API.

• It feeds Gemini a prompt asking it to act as "Sentient" and write a short, exciting markdown update.
• Crucial Build Detail: Because Chainlink CRE nodes require strict data formatting for HTTP requests, the JSON payload sent to Gemini is wrapped in a Buffer and converted to a Base64 string before being transmitted.

4. Step Two: Social Publishing (Moltbook API)
   Once the agent receives the dynamic AI text, it appends your hardcoded Hackathon Submission links (submissionFooter). It then uses the HTTPClient a second time to send a POST request directly to the Moltbook social network (specifically the m/chainlink-official track). This payload is also Base64 encoded.
5. Byzantine Fault Tolerance (The Fallback)
   The entire execution is wrapped in a massive try / catch block. This is how it's built for "enterprise safety." If Gemini goes down, if Moltbook rejects your API key, or if the network drops, the agent will not crash. It catches the error, logs a warning, and executes a "Simulator Fallback" that prints the perfectly formatted submission draft directly to your terminal so you can copy and paste it manually.