Spike
A quantum-safe, multi-skilled autonomous DeFi AI agent built on the 0G Network
Videos



Tech Stack
Description
Problem:
AI agents (single or multi-agent orchestrators) are no longer experimental. They already monitor markets 24/7, rebalance portfolios, optimize yields, hedge risks, execute trades, and manage liquidity positions.
Traditional ECDSA signatures (used on Ethereum, 0G's EVM layer, etc.) are vulnerable to future quantum attacks via Shor's algorithm.
Google Quantum AI research (March 2026) shows that breaking elliptic curve cryptography (like secp256k1 used in most blockchains) could require far fewer qubits (~500,000) and run in ~9 minutes
Solution:
Post-Quantum Cryptography
ML-DSA-65 (Dilithium3) (NIST FIPS 204) signs every agent action before it is written on-chain
ML-KEM-1024 (Kyber-1024) (NIST FIPS 203) + AES-256-GCM encrypts agent configs before upload to 0G Storage
Both algorithms run via
@noble/post-quantumin the browser — private keys never leave the client in plaintextKey fingerprints are registered on
PQCKeyRegistry.sol; the agent checks registration before executing any trade
On-Chain Skill Registry
SkillRegistrystores four DeFi skills on-chain: LP Provider, DCA Strategy, Lending/Borrowing, Simulated TradeUsers select a skill during the configure wizard; the
skillKey(keccak256 of skill ID) is passed todeployAgentAgentRegistryenforces that the skill is active inSkillRegistryat deploy time
Agent NFT (INFT)
Every deployed agent mints an ERC-721
AgentNFTthat stores the agent's Dilithium fingerprint and skill keytokenURIreturns live performance metadata: total actions, success rate, PnL bps, sourced fromAgentRegistryPQC-gated transfers: the receiving wallet must have a registered PQC key
Portfolio Optimization
services/quantumruns FastAPI + scipy locally — no external service requiredPortfolio weights use classical Markowitz (SLSQP) with an efficient frontier calculation
VaR-95 / CVaR-99 are estimated via Monte Carlo simulation (log-normal, 100k samples)
Rebalance trades are selected by a greedy cost-minimizing algorithm
Optimization results are cached in 0G Storage KV for 6 hours, keyed by portfolio hash — the agent skips re-running if drift is below
driftThreshold
0G Network Integration
0G Storage (
@0gfoundation/0g-storage-ts-sdk): stores Kyber-encrypted agent configs and appends immutable audit logs0G Compute (
@0glabs/0g-serving-broker): runsinferMarketRegimeto classify the current market as bull / bear / sideways / volatile (requiresZG_INFERENCE_PROVIDER)0G Chain (EVM, chainId 16602 testnet): five smart contracts govern key registration, attestation, skill catalog, agent NFTs, and lifecycle
Progress During Hackathon
- 5 contracts deployed & verified on 0G Testnet (chainId 16602) — PQCKeyRegistry, TeeAttestationVerifier, SkillRegistry, AgentNFT (ERC-7857 INFT), AgentRegistry; all source-verified on chainscan-galileo.
- Full 0G stack integrated — agent configs encrypted with ML-KEM-1024 and stored on 0G Storage, optimization results cached in 0G KV (6hr TTL), market regime classified via 0G Compute LLM, ML-DSA-65 signatures TEE-attested through 0G Compute TeeML and committed on-chain.
- Production deployment live — Next.js frontend on Netlify, FastAPI quantum optimizer on Railway, agent decision loop on Cloudflare Workers; TEE attestation wired end-to-end so deployAgent passes on the deployed version.
Fundraising Status
MVP status