Skip to main content

​
Key features

MetaVault AI combines cutting-edge AI agent technology with battle-tested DeFi protocols to deliver a vault that actively manages your capital.

​
Autonomous strategy monitoring

The Strategy Sentinel agent continuously monitors vault health without human intervention.

​
Real-time health checks

  • Strategy balances: Tracks deposited amounts, borrowed positions, and net exposure
  • Loan-to-Value (LTV) monitoring: Calculates current LTV ratio for leveraged positions
  • Liquidation risk detection: Identifies positions approaching danger zones
  • Price tracking: Fetches live LINK and WETH prices from CoinGecko API
// Strategy Sentinel monitoring loop
const states = await get_strategy_states();
const prices = await get_token_prices();
const leverageState = await get_leverage_strategy_state();

if (leverageState.ltv > 0.7) {
  // High risk detected - take action
  await toggle_leverage_strategy_pause();
  await auto_deleverage(5);
}

​
Intelligent alerting

The agent evaluates multiple risk factors:
  • LTV exceeding 70% threshold
  • Rapid price movements (>10% change)
  • Divergence from target allocations
  • Low buffer to liquidation threshold
  • Excessive leverage ratios
The Strategy Sentinel uses rule-based decision logic, not guesses. All actions are based on on-chain data and real market prices.

​
Dynamic risk management

MetaVault AI actively manages risk through multiple mechanisms.

​
Automatic deleveraging

When liquidation risk is detected, the system automatically reduces leverage:
  1. Withdraw LINK collateral from Aave
  2. Swap LINK → WETH via Uniswap
  3. Repay WETH debt
  4. Repeat until LTV is safe
function deleverageAll(uint256 maxLoops) external onlyRouter {
    for (uint256 i = 0; i < maxLoops; i++) {
        uint256 debt = pool.getUserDebt(address(this), WETH);
        if (debt == 0) break;
        
        // Calculate LINK needed (with 5% buffer for slippage)
        uint256 price = oracle.getPrice(WETH);
        uint256 linkNeeded = (debt * price * 105) / (1e18 * 100);
        
        // Unwind: withdraw → swap → repay
        pool.withdraw(address(token), linkNeeded, address(this));
        swapRouter.swapExactTokensForTokens(linkNeeded, 0, path, address(this), block.timestamp + 300);
        pool.repay(WETH, debt, 2, address(this));
    }
}

​
Leverage parameter tuning

The Strategy Sentinel can adjust leverage aggressiveness based on market conditions: Parameters:
  • maxDepth (1-6): Number of leverage loops
  • borrowFactor (0-8000 bps): Percentage to borrow per loop
  • paused (bool): Emergency pause flag
Adjustment strategy:
  • High volatility → reduce maxDepth and borrowFactor
  • Stable markets → maintain or increase parameters
  • Extreme risk → set paused = true
// Agent decision logic
if (priceVolatility > 0.15) {
  // Reduce leverage in volatile markets
  await update_leverage_params({
    maxDepth: 2,
    borrowFactor: 4000  // 40% instead of 60%
  });
} else if (ltv > 0.7) {
  // Emergency pause
  await toggle_leverage_strategy_pause();
}

​
Strategy pause mechanism

The leverage strategy can be paused instantly to prevent new risky investments: 
function invest(uint256 amount) external override onlyRouter {
    require(!paused, "paused");
    // ... investment logic
}

function togglePause() external onlyRouter {
    paused = !paused;
    emit PauseToggled(paused);
}

​
Intelligent portfolio rebalancing

MetaVault AI maintains optimal allocation across strategies based on risk profile and market conditions.

​
Target weight management

Default allocation:
  • Leverage strategy: 80% (high yield, high risk)
  • Aave safe strategy: 20% (stable yield, low risk)
The Strategy Sentinel can adjust these weights dynamically: 
await update_strategy_target_weights({
  leverageStrategy: 5000,  // 50%
  aaveStrategy: 5000       // 50%
});

await rebalance_vault();

​
Automatic rebalancing

The rebalance() function reallocates funds to match target weights:
  1. Calculate total managed assets
  2. Pull excess from overweight strategies
  3. Push funds to underweight strategies
  4. Update bookkeeping
function rebalance() external onlyOwner {
    uint256 totalManaged = _computeTotalManaged();
    
    // Pull from overweight strategies
    for (uint256 i = 0; i < strategies.length; i++) {
        address strat = strategies[i];
        uint256 current = IStrategy(strat).strategyBalance();
        uint256 desired = (totalManaged * targetBps[strat]) / 10000;
        
        if (current > desired) {
            uint256 excess = current - desired;
            IStrategy(strat).withdrawToVault(excess);
        }
    }
    
    // Push to underweight strategies
    uint256 vaultBal = vault.totalAssets();
    for (uint256 i = 0; i < strategies.length; i++) {
        address strat = strategies[i];
        uint256 current = IStrategy(strat).strategyBalance();
        uint256 desired = (totalManaged * targetBps[strat]) / 10000;
        
        if (current < desired && vaultBal > 0) {
            uint256 need = desired - current;
            uint256 amt = need <= vaultBal ? need : vaultBal;
            vault.moveToStrategy(strat, amt);
            IStrategy(strat).invest(amt);
            vaultBal -= amt;
        }
    }
}

​
Rebalancing triggers

The agent triggers rebalancing when:
  • Allocation drift exceeds 5%
  • Market conditions change significantly
  • New deposits increase vault balance
  • After updating target weights

​
Automated yield harvesting

The vault automatically compounds profits back into strategies.

​
Harvest mechanism

Each strategy implements a harvest() function: 
function harvest() external override onlyRouter {
    (address aTokenAddr, , ) = dataProvider.getReserveTokensAddresses(address(token));
    uint256 aBal = IERC20(aTokenAddr).balanceOf(address(this));
    
    if (aBal > deposited) {
        uint256 profit = aBal - deposited;
        uint256 out = pool.withdraw(address(token), profit, address(this));
        token.safeTransfer(vault, out);
        emit Harvested(out);
    }
}

​
Performance fee application

The vault takes a configurable performance fee on harvested profits: 
function handleHarvestProfit(uint256 profit) external {
    require(msg.sender == router, "not router");
    if (profit == 0) return;
    
    uint256 fee = (profit * performanceFeeBps) / 10000;
    if (fee > 0) {
        asset.safeTransfer(feeRecipient, fee);
    }
}

​
Harvest all strategies

The router can harvest all strategies in a single transaction: 
function harvestAll() external onlyOwner {
    for (uint256 i = 0; i < strategies.length; i++) {
        address strat = strategies[i];
        uint256 beforeBal = vault.totalAssets();
        
        IStrategy(strat).harvest();
        
        uint256 afterBal = vault.totalAssets();
        if (afterBal > beforeBal) {
            uint256 profit = afterBal - beforeBal;
            vault.handleHarvestProfit(profit);
            emit Harvested(strat, profit);
        }
    }
}

​
Natural language interface

The Chat agent provides an intuitive way to interact with the vault.

​
User-friendly commands

Users can chat naturally:
  • “What’s my balance?”
  • “Deposit 10 LINK”
  • “Show me the current APY”
  • “Withdraw all my funds”
  • “How much is LINK worth?”

​
Smart transaction preparation

The agent handles the complex approval flow:
  1. Check if user has approved LINK spending
  2. If not, prepare approval transaction
  3. After approval, prepare deposit transaction
  4. Return unsigned transactions for wallet signature
// Deposit flow in chat agent
const allowance = await check_allowance(userWallet, amount);

if (allowance < amount) {
  // Need approval first
  const approveTx = await approve_link(amount);
  return {
    reply: "Please sign this approval transaction to allow the vault to spend your LINK.",
    unsignedTx: approveTx,
    needsApproval: true,
    step: "approval"
  };
} else {
  // Already approved, prepare deposit
  const depositTx = await user_deposit(amount);
  return {
    reply: "Ready to deposit! Please sign this transaction.",
    unsignedTx: depositTx,
    needsApproval: false,
    step: "deposit"
  };
}

​
Privacy and security

The Chat agent enforces strict boundaries:
  • Only accesses data for the requesting user’s wallet
  • Never exposes admin functions (rebalance, harvest, risk parameters)
  • Cannot view other users’ balances or positions
  • Returns only public vault information (total assets, APY)
The Chat agent responds in JSON format for transactional operations, making it easy to integrate with frontend wallets.

​
Transparent performance tracking

Every user can track their individual performance.

​
User growth calculation

The vault maintains detailed accounting: 
mapping(address => uint256) public netDeposited;    // Total deposited
mapping(address => uint256) public totalWithdrawn;  // Total withdrawn

function userGrowth(address user) public view returns (int256) {
    uint256 deposited = netDeposited[user];
    if (deposited == 0) return 0;
    
    uint256 withdrawn = totalWithdrawn[user];
    uint256 currentValue = convertToAssets(balanceOf(user));
    
    // PnL = (current value + withdrawn) - deposited
    int256 pnl = int256(currentValue + withdrawn) - int256(deposited);
    return pnl;
}

function userGrowthPercent(address user) external view returns (int256) {
    uint256 deposited = netDeposited[user];
    if (deposited == 0) return 0;
    
    int256 pnl = userGrowth(user);
    return (pnl * 1e18) / int256(deposited);  // 1e18 = 100%
}

​
Share-based accounting

The vault uses ERC20 shares for fair profit distribution:
  • New deposits receive shares at current price
  • Yields increase total assets, raising share price
  • Withdrawals burn shares at current price
  • All users benefit proportionally from yields
function convertToShares(uint256 assets) public view returns (uint256) {
    uint256 s = totalSupply();
    return s == 0 ? assets : (assets * s) / totalManagedAssets();
}

function convertToAssets(uint256 shares) public view returns (uint256) {
    uint256 s = totalSupply();
    return s == 0 ? shares : (shares * totalManagedAssets()) / s;
}

​
Built with ADK-TS

MetaVault AI showcases the power of the Agent Development Kit for TypeScript.

​
Key ADK-TS features used

Conversation orchestration:
  • Session memory across interactions
  • Context state management for caching APY updates
  • Tool response formatting and guidance
TypeScript integration:
  • Fully typed agent logic and tools
  • Seamless Next.js/React integration
  • Developer-friendly experience
Modular architecture: 
import { AgentBuilder, LlmAgent } from "@iqai/adk";

const strategySentinel = new LlmAgent({
  name: "strategy_sentinel_agent",
  description: "Guardian & Portfolio Manager",
  instruction: "...",
  model: model,
  tools: [
    get_strategy_states,
    get_vault_state,
    rebalance_vault,
    auto_deleverage,
    // ...
  ]
});

const mainAgent = new AgentBuilder()
  .addAgent(strategySentinel)
  .addAgent(chatAgent)
  .build();

​
Benefits of ADK-TS

  1. Reduced complexity: Frontend doesn’t need complex state management
  2. Type safety: Compile-time checks for agent logic and tools
  3. Tool orchestration: Structured calling conventions and error handling
  4. Conversation memory: Agents remember context across messages
  5. Modular design: Easy to add new specialized agents

​
Error handling and safety

MetaVault AI includes comprehensive error handling.

​
Strategy operation safety

All strategy operations use try-catch blocks: 
try IStrategy(strat).invest(amount) {
    // success
} catch (bytes memory reason) {
    emit StrategyWithdrawFailed(strat, _decodeRevertReason(reason));
    // Continue to next strategy instead of reverting entire transaction
}

​
Withdrawal protection

The vault ensures users can always withdraw (if liquidity exists): 
function withdraw(uint256 shares) external returns (uint256 assetsOut) {
    assetsOut = convertToAssets(shares);
    _burn(msg.sender, shares);
    
    // Pull from strategies if vault doesn't have enough liquidity
    uint256 vaultBal = asset.balanceOf(address(this));
    if (vaultBal < assetsOut) {
        uint256 needed = assetsOut - vaultBal;
        IStrategyRouter(router).withdrawFromStrategies(needed);
        require(
            asset.balanceOf(address(this)) >= assetsOut,
            "not enough liquidity after pull"
        );
    }
    
    asset.safeTransfer(msg.sender, assetsOut);
}

​
Leverage safety mechanisms

The leverage strategy includes multiple safeguards:
  • Maximum depth capped at 6 iterations
  • Borrow factor limited to 80% (8000 bps)
  • Conservative borrowing based on pool liquidity
  • Automatic loop breaking on swap failures
  • Try-catch on all Aave interactions

​
Next steps

Getting started

Set up and run MetaVault AI locally

API reference

Explore smart contract interfaces and agent tools

Build docs developers (and LLMs) love

Get started for freeTalk to us