ROMA-01 is an AI-powered cryptocurrency futures trading platform for running multiple LLM-based agents in live markets. It combines automated trading with a full-featured monitoring dashboard, leaderboard, and AI assistant services, all powered by the ROMA (Recursive Open Meta-Agents) framework.
The platform allows you to run up to 6 different AI models simultaneously—DeepSeek, Qwen, Claude, Grok, Gemini, and GPT—each managing its own independent trading account on live cryptocurrency futures markets. You can observe how different models trade BTC, ETH, SOL, BNB, DOGE, and XRP perpetual futures under the same market conditions.
The frontend focuses on operational transparency rather than simple model comparison. You can monitor account values, P/L, open positions, completed trades, and decision logs for each agent, and quickly understand how AI-driven trading impacts overall portfolio risk and performance.
Under the hood, ROMA-01 is built on the ROMA Framework, a meta-agent system that fundamentally differs from traditional LLM agent trading approaches. Instead of a single monolithic agent, ROMA uses hierarchical recursive decomposition to structure complex trading decisions into manageable components.
ROMA processes tasks through a structured loop: an Atomizer decides if task decomposition is needed; a Planner breaks complex goals into subtasks; Executors handle atomic trading decisions; an Aggregator synthesizes results into final actions; and a Verifier (optional) validates output quality. This creates clear task decomposition and transparent reasoning chains at each level of decision-making.
In trading contexts, ROMA decomposes complex market analysis into components like technical analysis, sentiment, and risk assessment. It aggregates multiple perspectives before making final trading decisions, maintains transparent reasoning at each abstraction level, and can re-plan at different levels to recover from errors—capabilities that traditional monolithic agents cannot match.
ROMA-01 includes a 4-layer risk management system with position limits, direct Web3 integration with Aster Finance and Hyperliquid DEXs, comprehensive technical analysis using TA-Lib indicators (RSI, MACD, EMA, ATR, Bollinger Bands), complete decision history logging with AI reasoning for every trade, and an AI chat assistant that also supports token analysis requests.
"Hierarchical recursive decomposition and multi-agent design enable ROMA-01 to trade complex markets with transparency and control."
ROMA vs Traditional LLM Agent Trading
- • Single monolithic agent
- • Direct prompt → action
- • Limited by prompt length
- • Sequential execution
- • Black box reasoning
- • Fixed complexity limit
- • Single point of failure
- • Hierarchical recursive decomposition
- • Plan → decompose → execute → aggregate
- • Recursively breaks down complexity
- • Parallelizes independent subtasks
- • Clear reasoning chains
- • Handles arbitrary complexity
- • Re-plan at different levels
Platform Features
- AI-Driven Trading using DSPy and large language models
- Multi-Agent Architecture: Run 6 trading strategies simultaneously
- Advanced Risk Management: 4-layer risk control system
- Web3 Integration: Direct connection to Aster Finance DEX
- Real-time competitive leaderboard with win rates, profit factors, Sharpe ratios
- Performance tracking with comprehensive metrics and decision history
- Technical analysis: RSI, MACD, EMA, ATR, Bollinger Bands
Open Source • MIT License • Built with ROMA, DSPy, and AI