AXIOM is a high-fidelity simulation platform where dozens of autonomous Reinforcement Learning (RL) agents trade, form coalitions, and exhibit emergent market behaviors—without being programmed with the rules of economics.
"You build the world; they invent the economy."
Most trading systems are reactive. AXIOM agents are reasoning, negotiating, and self-organizing. It is a living laboratory for emergent intelligence, designed for research in multi-agent systems and computational economics.
- Heterogeneous Agents: Agents with varying utility functions (CRRA) and risk tolerances.
- Emergent Price Discovery: Market prices are derived from agent interactions, not hard-coded formulas.
- Causal Policy Injection: A "God Mode" sandbox to test economic interventions (taxes, shocks) mid-simulation.
- Behavioral Phase Transitions: Observe cooperation vs. defection under resource stress.
- Engine: Mesa (Agent-Based Modeling)
- Intelligence: Ray RLlib, PyTorch (PPO Algorithm)
- Analytics: NetworkX, Pandas, NumPy
- Visualization: Plotly Dash (Real-time), D3.js (Network Graphs)
- API: FastAPI
- Phase 1: Simulation Foundation — LOB, Double Auction, Walrasian clearing.
- Phase 2: RL Intelligence Layer — Integration with Ray RLlib & Multi-Agent training.
- Phase 3: Emergent Complexity — Coalition formation & Game Theoretic signaling.
- Phase 4: Research-Grade Polish — D3.js visualizations & statistical reporting.
pip install "mesa<3.0" ray[rllib] torch networkx fastapi uvicorn dash pandas numpy plotlypython run_sim.pypython axiom/dashboard/app.pyView the live market at http://127.0.0.1:8050.
$env:PYTHONPATH="."
python axiom/rl/train.py- Emergent price convergence in zero-intelligence vs. RL-based markets.
- Stability of coalitions under exogenous supply shocks.
- Impact of wealth redistribution policies on Gini coefficient dynamics.