Compose trading strategies as scores. Rehearse against live data. Conduct onchain with our MPC-secured smart accounts and ampli's onchain policy engine.
score TreasuryOptimizer { // Which protocols we conduct through ensemble: { lender: @aave_v3 alt: @morpho_blue dex: @uniswap_v3 } // Guardrails enforced at runtime dynamics: { max_single_move: 500000 USDC max_slippage: 50bps require_approval_above: 250000 USDC } cadenzas: { risk: { model: "anthropic:claude-opus-4-6" timeout: 15 fallback: false } } on hourly: { spread = apy(@alt, USDC) - apy(@lender, USDC) if spread > 100bps { decision = cadenza risk: "Spread is {spread}bps. Rebalance?" { output: { type: boolean } fallback: false } if decision { measure rebalance: { lender.withdraw(USDC, balance(USDC)) alt.lend(USDC, balance(USDC)) } } } } }
A single score flows from agent to SDK, fans out across instruments, and lands onchain. Every note is a real pipeline step.
Every score follows the same safety pipeline. Nothing moves onchain until you've rehearsed it and reviewed the feedback.
Scores run on Ampli's full infrastructure stack. MPC custody, onchain governance, smart accounts, and policy enforcement that cannot be bypassed.
No single party holds a complete key. The Ampli MPC distributes key shares between client and server. The full private key is never reconstructed, even during signing.
Ampli's policy engine enforces rules at the smart contract level. Even a compromised backend cannot bypass validation. Agent identity, whitelists, and roles live onchain.
ERC-4337 account abstraction with ERC-7579 modular validation. Gas sponsorship, batched operations, programmable validation, and deterministic addresses.
AI decision points with schema validation, timeouts, and fallback values. The model provides judgment; the score controls what happens with it. Fully auditable and replayable.
Every execution produces a typed receipt before settlement. Drift detection at commit time rejects stale previews. Rehearse as many times as needed before conducting live.
Enterprise Message Queue distributed workers process scores across wallets in parallel. One scheduler, N stateless workers, horizontally scalable execution independent of the API.
Each instrument has an adapter that translates score actions into protocol-specific calls. Switch instruments without rewriting strategy logic.
| Instrument | Type | Actions | Chains |
|---|---|---|---|
| @aave_v3 | Lending | lend, withdraw, borrow, repay | Ethereum, Base |
| @morpho_blue | Lending | lend, withdraw, supply/withdraw collateral | Base |
| @uniswap_v3 | DEX | swap | Multi-chain |
| @uniswap_v4 | DEX | swap | Multi-chain |
| @across | Bridge | bridge | Cross-chain |
| @fluid | Lending | lend, withdraw | Ethereum, Base |
| @pendle | Yield | add/remove liquidity, market ops | Multi-chain |
Write scores locally or send intents through the Platform API. Both run the same conductor underneath.
Compose once. Rehearse as many times as you need. Conduct onchain with our MPC-secured smart accounts and ampli's onchain policy engine.