YieldPilot

AI mandate-driven stablecoin yield copilot. Built for the DeFi Mullet Hackathon #1 (AI x Earn track).

Live: https://yieldpilot-iota.vercel.app

What It Does

YieldPilot does not just find the highest yield. It enforces your rules.

1. Connect your wallet
2. App scans your USDC/USDT balances across Ethereum, Base, Arbitrum, and Optimism
3. Describe your strategy in plain English or choose a preset mandate
4. AI translates that strategy into a structured mandate with TVL, break-even, chain, and reward constraints
5. App filters 659+ vaults against that mandate
6. App fetches a Composer quote and calculates whether the move is economically worth it
7. If the move passes your mandate and breaks even fast enough: execute
8. If not: explicitly refuse and explain why your funds stay put
9. App shows a visible decision trace from prompt to final verdict

The core differentiator is the combination of:

• natural-language mandate parsing
• visible AI decision trace
• deterministic vault ranking
• real Composer cost analysis
• refusal logic when the move is not worth it

How It Uses the Earn API

Earn Data API (earn.li.fi)

• GET /v1/vaults with full pagination across all 659+ vaults
• GET /v1/chains for supported chain metadata
• GET /v1/protocols for protocol metadata and maturity scoring
• All requests are authed with x-lifi-api-key (same key as the Composer routes)
• Proxied through GET /api/earn in the app so the browser does not depend on direct third-party API access

Composer (li.quest)

• GET /v1/quote to build deposit routes and extract gas/fee cost data
• GET /v1/status to track execution progress after deposit
• Cost data feeds the break-even analysis engine

On-Chain Reads

• viem readContract for USDC/USDT balances across 4 chains (Ethereum, Base, Arbitrum, Optimism)

AI Intent Layer

• POST /api/intent converts a natural-language strategy into a custom mandate
• Prefers Bankr LLM Gateway when BANKR_LLM_KEY is present
• Falls back to OpenAI when OPENAI_API_KEY is present
• Falls back to a deterministic parser when no model is configured
• AI output directly changes:
  • cross-chain allowance
  • TVL floor
  • break-even ceiling
  • reward-heavy exclusions
  • APY improvement floor

Mandate System

Three presets remain, but they are no longer the only path.

Users can still choose:

Mandate	TVL Floor	Max Break-Even	Cross-Chain	Reward-Heavy
Conservative	$100M	7 days	No	Excluded
Balanced	$10M	14 days	Allowed	Allowed
Aggressive	$1M	30 days	Allowed	Allowed

Or they can type a strategy like:

• "Keep me on the safest same-chain vault and do not bridge if payback is longer than 7 days."
• "Find me the best stablecoin yield above 5% even if it needs a bridge, but keep TVL above $50M."

That prompt becomes a custom typed mandate.

Break-Even Engine

Before recommending any move, YieldPilot:

1. Fetches a real Composer quote for the route
2. Extracts total gas + fee cost in USD
3. Calculates APY delta vs idle stablecoins at 0%
4. Computes daily incremental yield
5. Derives break-even days = route cost / daily yield improvement
6. Compares against mandate threshold

If break-even exceeds the mandate limit, the move is refused with a clear explanation.

Ranking Engine

Vault selection stays deterministic. Factors:

• APY level (0-30 points)
• APY stability: 1d vs 30d drift (0-20 points)
• TVL depth (0-20 points)
• Protocol maturity tier (0-15 points)
• Same-chain bonus (0-15 points)

AI is used for intent translation, not for arbitrary vault picking. That means:

• AI converts vague user strategy into a strict mandate
• deterministic ranking selects the vault
• Composer verifies whether execution is economically worth it

This keeps AI load-bearing without making selection non-reproducible.

Tech Stack

• Next.js 16 + TypeScript
• wagmi + RainbowKit for wallet connection
• viem for on-chain reads and transaction building
• Tailwind CSS 4
• LI.FI Earn Data API + Composer
• Bankr LLM Gateway / OpenAI API (optional, for natural-language mandates)
• Vercel for deployment

Quick Start

git clone https://github.com/dolepee/yieldpilot.git
cd yieldpilot
npm install
cp .env.local.example .env.local  # add your LIFI_API_KEY
npm run dev

Environment Variables

Variable	Description
LIFI_API_KEY	Composer API key from portal.li.fi
NEXT_PUBLIC_WALLETCONNECT_PROJECT_ID	WalletConnect project ID
BANKR_LLM_KEY	Optional. Enables Bankr-backed natural-language mandate parsing
BANKR_LLM_MODEL	Optional. Bankr model name, defaults to deepseek-v3.2
OPENAI_API_KEY	Optional. Enables model-backed natural-language mandate parsing

Project Structure

src/
  app/
    page.tsx            # Main app flow
    providers.tsx       # wagmi + RainbowKit
    api/quote/route.ts  # Server-side Composer proxy
  components/
    Header.tsx          # Nav + wallet connect
    WalletSnapshot.tsx  # Stablecoin balance display
    MandatePicker.tsx   # Mandate selection UI
    RecommendationCard.tsx  # Top vault recommendation
    WorthItCard.tsx     # Break-even analysis verdict
    ExecutionTracker.tsx    # Tx status polling
    YieldStoryCard.tsx  # Shareable result card
    EarnStats.tsx       # Vault/chain/APY stats
  hooks/
    useStablecoinBalances.ts  # Multi-chain balance reads
    useEarnData.ts      # Vault/chain/protocol fetch
    useLifiStatus.ts    # Transaction status polling
  lib/
    constants.ts        # Addresses, tiers, config
    earn-api.ts         # Earn Data API client
    composer.ts         # Composer quote client
    mandates.ts         # Mandate types and presets
    ranking.ts          # Deterministic vault ranking
    worth-it.ts         # Break-even analysis
    demo.ts             # Demo mode mock data
    wagmi.ts            # Wallet config

Hackathon

Built for the DeFi Mullet Hackathon #1 by LI.FI.

• Track: AI x Earn
• Duration: April 8-14, 2026
• Prize pool: $5,000 USDC

License

MIT