
RuthAgent is a fully autonomous crypto trading agent that runs in a continuous loop, combining live market intelligence with research-backed AI reasoning and on-chain trust infrastructure. ARCHITECTURE Each 15-second cycle executes: Market Data → Signal Processing → LLM Decision → Paper Trade → On-Chain Attestation. The agent fetches live BTC/USD and ETH/USD prices and 1-hour OHLC candles from Kraken CLI, pre-processes them into structured signals (trend direction, momentum, volatility regime, volume surge, support/resistance levels), and feeds these to Groq's llama-3.3-70b-versatile model for strategic trade decisions. RESEARCH-BACKED DESIGN Informed by three key papers: 1. AI-Trader (arXiv:2512.10971) — Risk control outperforms raw intelligence for cross-market robustness. RuthAgent uses a risk-first prompt that forces the LLM to classify market regime and assess risk budget before deciding. 2. Inter-Agent Trust Models (arXiv:2511.03434) — No single trust mechanism is sufficient. RuthAgent layers identity (ERC-8004 NFT), reputation (cross-agent feedback), and validation (checkpoint attestations) for multi-layered trust. 3. Agentic RAG (arXiv:2501.09136) — Minimal curated context outperforms raw data injection. The agent pre-processes OHLC candles into compact signals locally, giving the LLM clean directional inputs rather than raw arrays. ON-CHAIN TRUST (ERC-8004) - Registered identity on Sepolia (Agent ID 89) - Posts validation attestations via postAttestation after every trade cycle - Submits EIP-712 signed trade intents to the RiskRouter - Provides cross-agent reputation feedback to peer agents via the hackathon ReputationRegistry - Claims vault capital for economic skin-in-the-game EXECUTION Paper trades execute through Kraken CLI with automatic position management. The agent handles both BUY and SELL across BTC/USD and ETH/USD.
12 Apr 2026