The problem AgentFi solves

"What if your AI agent could take out a loan, trade, and repay with profit—all without you lifting a finger?"

AgentFi is a multi-agent P2P lending marketplace where AI agents autonomously borrow capital, deploy it via quantitative trading strategies, and repay with on-chain reputation. We solve the #1 problem in DeFi: Capital Inefficiency.

The Problem

1. DeFi Lending is Dead Capital: Over-collateralized protocols (like Aave) lock up $3 for every $1 borrowed.
2. AI Agents are "Broke": Existing frameworks assume agents already have funds; there is zero infrastructure for agents to access unsecured capital.
3. Sybil-Prone Reputation: Most reputation systems are easily gamed by wash-trading.

Our Solution

AgentFi creates an Agent-Native Credit Market:

• ZK-Verified Identity: Using Reclaim Protocol & ENS to ensure "One Human = One Identity Ecosystem" (Anti-Sybil).
• Continuous Collateral Curve: No binary liquidation cliffs. Collateral requirements drop from 100% to 0% as agents prove their discipline.
• Autonomous Execution: 2-of-2 multisig wallets allow agents to trade 24/7 without manual user signing for small trades (<$500/hr).
• Reputation as Collateral: On-chain reputation (via PeosFi) becomes the "Working Capital" for the agent economy.

Key Innovations (The "Top 1%" Tech)

• The Identity Stack: ZK-Proof (Reclaim) + ENS + userId isolation ensures it is mathematically impossible to benefit from wash-trading.
• Credit-Score Model: A 0–50 scale that excludes trading "Alpha" (skill) and focuses strictly on "Repayment Discipline" (risk).
• The Matcher Engine: A sophisticated P2P orderbook where lenders compete on interest rates to fund high-reputation agents.

⁉️ FAQs

Q: "How do you prevent a total collapse on zero-collateral loans?"

A: We use a Continuous Collateral Curve. Agents don't get zero-collateral access until they reach Reputation 35, which requires 5+ successful trade/repay cycles. Our crossover point is modeled on an 8% default rate (the DeFi benchmark for unsecured loans). We don't lend to code; we lend to provable discipline.

Q: "Why exclude trading profit (Alpha) from the Reputation score?"

A: High-alpha strategies are often high-risk. Rewarding alpha with more capital compounds systemic risk. Instead, alpha is surfaced to lenders so they can price their interest rates accordingly. Profitability is a pricing signal; repayment is a trust signal.

Q: "Is the 2-of-2 Multisig decentralized?"

A: We use the multisig as an Autonomous Risk Gate. The platform co-signs trades <$500/hr instantly to enable 24/7 autonomy. Anything larger requires user approval. This moves the human from the execution path (slow) to the policy path (safe).

🫦 Tech Stack

• Blockchain: Base Sepolia (Lending Contracts, MockUSDC)
• Identity: ENS (Ethereum Sepolia) + Reclaim Protocol (ZK-Proof)
• Wallet: Metamask
• Trading: HeyElsa API (Portfolio & Swap Execution)
• Storage: Fileverse (Encrypted Strategy Docs)
• Backend: Node.js, Express, PostgreSQL, Redis

🏁 Impact

AgentFi is more than a tool—it's the financial infrastructure for the agentic economy. As billions of agents enter the web, AgentFi provides the bank accounts and credit scores they need to be truly self-sovereign.

Challenges we ran into

Challenges we ran into 🏃‍♂️

Cross-Chain Identity Resolution: Our lending logic lives on Base Sepolia, but ENS names are registered on Ethereum Sepolia. Resolving these identities across chains in real-time created significant latency issues. We overcame this by implementing a dual-provider listener in our backend that caches ENS-to-Wallet mappings, ensuring agent verification happens in milliseconds without blocking the matching engine.

The "Double-Spend" vs. Autonomy Paradox: Implementing a 2-of-2 multisig for an autonomous agent presented a race-condition risk: the agent and the platform co-signer could both attempt to sign different transactions simultaneously. We solved this by building a Synchronous Transaction Queue in Redis, ensuring that the agent's rolling 60-minute volume gate is calculated and locked before the platform co-signs any transaction.

ZK-Proof Mapping & Privacy: Integrating Reclaim Protocol to prevent sybil attacks while maintaining user privacy was difficult. We had to derive a deterministic but anonymous humanId from the ZK proof. We implemented a hashing salt stored on-chain to ensure that a single human cannot register multiple ENS names, without ever exposing the user's underlying credentials to our database.

ZK-Proof Mapping & Privacy: Executing complex portfolio swaps via API during volatile testnet conditions often led to transaction slippage. We developed a "Pre-flight Verification" service that checks liquidity on Base before the agent requests a loan, preventing agents from borrowing capital for strategies that cannot be executed due to market depth.