TrustTrade AI — Multi-Agent Crypto Trading System
TrustTrade AI is a multi-agent autonomous cryptocurrency trading system designed for both the Kraken CLI and ERC-8004 hackathon tracks. Architecture: Five specialized agents collaborate in a pipeline — Signal Agent analyzes markets using trend (MA crossover) and momentum (RSI + MACD) strategies; Risk Agent enforces drawdown limits, daily loss caps, and volatility checks; Portfolio Agent calculates position sizing with dynamic stop-loss and take-profit (ATR-based); Manager Agent makes the final decision and generates ERC-8004 validation artifacts; Execution Agent routes orders to Kraken CLI (paper/live) or internal simulation. Fundamental Intelligence: Three additional agents enrich signals — Funding Rate Agent monitors perpetual swap funding, Sentiment Agent aggregates Fear & Greed Index and crypto news headlines, and On-Chain Agent tracks mempool activity, fees, and hashrate changes. ERC-8004 Integration: Every trade is cryptographically validated through three on-chain registries — Identity Registry (agent NFT registration on Base Sepolia), Validation Registry (EIP-712 signed trade intents), and Reputation Registry (trading yield and success rate feedback). This creates a fully auditable, trustless record of AI trading decisions. Key Results: The balanced preset achieves +2.3% return with 66.7% win rate and +2.31pp alpha over buy-and-hold on 30-day BTC/USD backtests. Five configurable style presets (balanced, aggressive, conservative, trend-following, mean-reversion) serve different risk profiles. An interactive Streamlit dashboard provides real-time monitoring with bilingual support (EN/中文). Tech Stack: Python, Streamlit, Kraken CLI, Web3.py, Base Sepolia (ERC-8004), Claude API (optional LLM Manager).
Category tags:"good use of agentic technologies with use of subagents too"
Dharma Singh
Senior Development Manager
"1. Application of Technology: 4.5 / 5 Justification: Very high execution. Combining the Python/Streamlit data architecture with Web3.py for Base Sepolia communication, plus the Kraken CLI execution layer, requires threading together three very different tech stacks beautifully. 2. Presentation: 4.5 / 5 Justification: Excellent. The bilingual (EN/zh) Streamlit dashboard makes this highly accessible to a global audience. The inclusion of the "5 Configurable Style Presets" (aggressive, conservative, trend-following, etc.) makes the demo very easy to digest for non-technical users. 3. Business Value: 4 / 5 Justification: High value for retail automation. While not as institution-focused as CrossMind or JudyAI's "capital preservation" pitches, giving retail users a dashboard with preset risk tolerances (like a robo-advisor) while guaranteeing on-chain transparency is a highly marketable product. 4. Originality: 3.5 / 5 Justification: The multi-agent layout (Signal -> Risk -> Portfolio) is the most standard architectural pattern for these hackathon entries. While executed exceptionally well, it doesn't introduce a radically new concept like "Proof of Preservation" or "Walk-Forward Validation." ⚖️ Pros & Cons Pros: Highly Accessible: The bilingual Streamlit dashboard and "Robo-advisor style presets" (Aggressive, Balanced, Mean-Reversion) make it very easy for a retail user to deploy their own AI agent. Comprehensive Data: Providing 3 distinct fundamental intelligence agents (Funding Rate, Sentiment/Fear & Greed, and On-Chain mempool tracking) gives the Signal Agent much better context than just staring at an RSI line. Full Integration: Successfully implements both the Kraken CLI execution requirement and the ERC-8004 Base Sepolia validation requirement. Cons: Standard Logic: The underlying strategies (MA crossover, RSI, MACD) are the most generic setups in all of algorithmic trading. Centralized "Manager": It relies on a single "Manager Agent" to resolve conflicts from the other agents, which could become a single point of failure (hallucination) compared to architectures that require strict consensus to execute."
Sanem Avcil
For Innovators & Creators