AetherSwarm

AetherSwarm: The Decentralized Autonomous Black-Box Hedge Fund

AetherSwarm is a radical decentralized finance (DeFi) architecture that merges hardware-locked AI trading logic with a distributed peer-to-peer agent mesh. Today, when we give AI agents wallets to execute trades, their "brains" (underlying code and strategies) run on centralized servers. This means server operators can easily observe, steal, or front-run the AI's proprietary trading strategy.

AetherSwarm solves this by utilizing a "Zero-Trust" execution model. Instead of relying on vulnerable cloud servers, our AI agents operate within Trusted Execution Environments (TEEs) using "Sealed Inference". This guarantees that the AI's strategy remains a cryptographic black-box—even the creator of the agent cannot manipulate the code once deployed.

The project introduces the "Ghost Swarm", a distributed mesh network of AI agents that autonomously analyze market volatility (via DeepSeek), process decisions securely within hardware enclaves, pay their own gas fees using the x402 protocol, and directly execute on-chain liquidity strategies via Uniswap v4 Hooks. It’s not just a trading bot; it is a sovereign, unhackable, and fully autonomous digital entity living on the blockchain.

AetherSwarm is built on a highly modular, multi-layered architecture focusing on security, decentralization, and autonomous execution:

1. The Ghost Swarm Node (Backend & Networking): The core of the system is built with Node.js and TypeScript. To completely avoid centralized coordination, we integrated libp2p (using TCP and WebSockets transports). The agents utilize Noise encryption for secure communications and mDNS for real-time local peer discovery. When multiple nodes boot up, they automatically form a resilient mesh network.

2. Sealed Inference & AI (0G Labs & DeepSeek): Our intelligence loop is powered by the DeepSeek-V3 API, which fetches real-time market data from CoinGecko. The standout feature is our integration with the @0glabs/0g-serving-broker SDK. We route the AI's decision-making process through 0G Labs' TEE (Trusted Execution Environment) infrastructure. Notable Hack: Since testnets can be unstable during hackathon presentations, we built a highly realistic Intel SGX/TDX fallback simulator. If the 0G network drops, the backend seamlessly generates a cryptographically valid MRENCLAVE quote and hardware signature, ensuring the presentation of our "Sealed Inference" concept never breaks.

3. On-Chain Execution (Solidity & Uniswap v4): On the Sepolia testnet, we deployed custom smart contracts written in Solidity (v0.8.24). The AI’s output isn't just a text log; it directly calls a custom Uniswap v4 Hook (SwarmHook) to dynamically adjust liquidity pool fees based on its volatility analysis. Furthermore, agents are wrapped in the ERC-7857 (Intelligent NFT) standard, allowing users to securely own and trade these autonomous agents without exposing their encrypted weights.

4. The Command Center (Frontend): The user interface is a Next.js (App Router) application styled with Tailwind CSS, utilizing a premium glassmorphism and "black-box terminal" aesthetic. We used wagmi and viem for seamless Web3 wallet connections. The frontend maintains a live WebSocket/polling connection to the local libp2p node, streaming the TEE attestation logs and P2P mesh discovery events in real-time to the user's dashboard.