The Operating System
System R AI is a 10 layer trading operating system for human traders and autonomous AI agents.
It grew from a single empirical observation: return volatility that exceeds system risk containment capacity produces terminal drawdowns. This was not a theory. It was a 1.7M unrealised and ended in complete destruction. The mathematical relationship extracted from that cycle became the foundation of the entire architecture.
What It Is
A complete financial operating system. Not a wrapper around a language model. Not a dashboard. Not a set of APIs bolted together. A unified architecture where every layer knows about every other layer, and risk governance is enforced at the infrastructure level before any order reaches a broker.
10 operational layers: identity, intelligence orchestration, probabilistic risk management, planning, execution routing, data persistence, analysis, memory, compliance, and operations.
187 domain services. 55 MCP tools. 25 broker and exchange adapters. 384 API endpoints. 13,241 verified tests. Every financial value in exact decimal precision.
Two Species, One System
Humans interact through NEO, a conversational trading interface. NEO operates the entire workflow: intelligence, planning, execution, risk, memory. The human sets the rules. NEO enforces them.
AI agents connect through MCP protocol or Python SDK. Three lines of code to first pre trade risk gate. Per call compute credits, no subscriptions.
Both species consume the same compute infrastructure. Both are governed by the same probabilistic risk architecture. The system does not distinguish between human intention and agent instruction. It validates both against the same mathematical constraints.
Model Agnostic
The system supports every qualifying language model. Claude, GPT, Llama, Mistral, DeepSeek, Gemini. The user chooses, or the balancer routes by task qualification.
The value is not in the language model. Language models are the interface layer. The value is in the 187 domain services that sit beneath: G Score position sizing, Iron Fist risk validation, Monte Carlo simulation, Kelly criterion, regime detection, cognitive bias detection, behavioral fingerprinting, trajectory prediction, and complete audit trail at every layer.
What It Provides
Risk management. Iron Fist validation enforced at infrastructure level. Kill switch, position limits, portfolio risk checks, daily loss controls. Every order passes through risk before reaching a broker.
Memory. Agents store observations, recall patterns, and detect their own cognitive biases. Behavioral fingerprinting and trajectory prediction improve with every session.
Markets. 9 asset classes with dedicated risk, sizing, and intelligence services. Equities, options, futures, forex, crypto, prediction markets. 25 brokers including IBKR, Schwab, Binance, Coinbase, Kraken, OANDA, dYdX, Polymarket.
Security. Per agent AES 256 encryption across all data paths. Strategies, credentials, and wallet data encrypted at rest. Only the owning agent can decrypt.
Compliance. Complete audit trail at every layer. What data was read, what checks passed, what was decided and why. When the regulator asks, the answer exists.
The Research Foundation
Every framework in System R AI was first tested through direct trading, then documented in the Principles of Trading research series, then encoded into the platform as executable infrastructure.
Position sizing as the primary driver of outcomes. Volatility as information rather than noise. Asymmetry as the foundation of edge. Feedback loops as the mechanism of adaptation. These are not features. They are the architecture itself.
The research came first. The system came from the research. The token came from the system.
Links
systemr.ai · agents.systemr.ai · docs.systemr.ai · GitHub
Explore: OSR Protocol · Work · Research · About