System R: The AI Trading Operating System

System R is the AI trading operating system built by System R AI. It connects 187 domain services, 55 MCP tools, and 25 broker adapters behind a single conversational interface. Traders talk to System R the way they would talk to an analyst. Ask a question, get an answer backed by real computation. Execute a trade, get a confirmation block with risk checks already applied.

What System R Is

System R is not a dashboard. It is not a screener. It is not a charting platform with an AI chatbot bolted on.

It is an operating system for systematic trading.

The difference matters. A dashboard shows you data. A screener filters instruments. A charting tool visualizes price. Each solves one problem, and you stitch them together yourself. You are the integration layer. You copy numbers between tabs, cross-reference tools, and make decisions while context-switching between six browser windows.

An operating system unifies everything. Market data, risk analysis, backtesting, position sizing, execution, and portfolio monitoring all live in one environment. The AI does not just answer questions. It operates within the full stack. When you ask "What is my portfolio heat right now?" the answer comes from the same risk engine that enforces your drawdown limits.

This is the core design principle behind System R. Every component is connected. Every tool call draws from the same domain layer. Nothing is bolted on.

The Chat Interface

System R's interface is conversational. Inspired by Claude.ai, it uses a chat-first design: sidebar for conversation history, a composer for input, and rich response blocks that go beyond plain text.

The decision to build chat-first was deliberate. Dashboards assume you know what to look at. Screeners assume you know what to filter for. A conversational interface assumes nothing. You state what you need, and the system figures out which of its 187 services to invoke. New users get the same analytical depth as experienced ones. The interface adapts to the question, not the other way around.

Rich Blocks

Responses in System R are not walls of text. They render as structured blocks:

Block Type	What It Shows
Text	Analysis, explanations, reasoning
Code	Strategy code, configuration snippets
Table	Market data, comparison matrices, portfolio breakdowns
Stat Grid	Key metrics at a glance (G-Score, Sharpe, drawdown, win rate)
Confirmation	Trade proposals requiring human approval before execution
Status	System state, order fills, position updates
Alert	Risk warnings, drawdown notifications, kill switch triggers

When System R proposes a trade, it does not just say "buy 100 shares of AAPL." It renders a confirmation block showing the order details, the risk impact on your portfolio, the position sizing calculation, and the current portfolio heat level. You approve or reject with a single click.

Model Selector

System R is model-agnostic. Nine LLMs are available, and you choose which one drives each conversation. The options span the cost-capability spectrum:

• High capability, higher cost. Claude Opus, GPT-4o. Deep reasoning, complex analysis.
• Balanced. Claude Sonnet, GPT-4o-mini. Good for most daily tasks.
• Fast and affordable. Claude Haiku, Llama models. Quick lookups, simple queries.

You can switch models mid-session. Start a conversation with a fast model for data retrieval, then switch to a reasoning-heavy model for strategy analysis. The context carries over.

System R's value is not the LLM. The LLM is the interface layer. System R's value is the domain layer underneath it: the G-Score, the risk engine, the 187 services. Any model that supports tool calling can access that layer. Users choose based on their needs, not ours.

The Domain Layer

This is where System R separates from everything else on the market.

187 domain services encode a decade of systematic trading knowledge into executable code. These are not API wrappers. They are purpose-built services covering five core areas:

Risk engine. G-Score calculation, drawdown analysis, portfolio heat monitoring, correlation tracking, VaR computation, stress testing against historical crisis periods. The risk engine runs continuously, not on demand.

Market analysis. Regime detection classifies current conditions as trending, mean-reverting, volatile, or quiet. Volatility modeling, liquidity assessment, and market structure analysis inform every decision.

Strategy framework. Backtest engine with walk-forward optimization. Monte Carlo simulation across thousands of trade-order permutations. Parameter sensitivity analysis to detect overfitting before it costs real money.

Capital management. Position sizing using Kelly, fractional Kelly, and risk-based methods. Equity curve management and reserve allocation. The 90% rule is built into the default sizing logic.

Execution. Order management, smart routing across brokers, fill analysis, slippage tracking. Every order passes through a risk gate before reaching any broker.

The domain layer is what makes System R a trading operating system rather than a chat interface. Strip away the LLM and the domain layer still functions. It is the core.

Connected Markets

System R connects to 25 brokers and exchanges across six asset classes. Every adapter is a real implementation, averaging 800+ lines of production code. These are not mock APIs or thin wrappers.

Asset Class	Example Brokers/Exchanges	Coverage
Equities	Interactive Brokers, Alpaca, Tradier	US, international
Options	Interactive Brokers, TD Ameritrade	US equity options
Futures	Interactive Brokers, CQG	CME, ICE, Eurex
Forex	OANDA, Interactive Brokers	Major and minor pairs
Crypto	Binance, Coinbase, Kraken, Bybit	Spot and perpetuals
Commodities	Interactive Brokers	Energy, metals, agriculture

The agent does not need to know which broker is connected. It calls market_data("AAPL") or submit_order("BTC-USD", "buy", 0.5) and the platform routes to the correct adapter. Broker-specific authentication, order normalization, and error handling are abstracted away.

A single ATOM workspace can connect to multiple brokers simultaneously. Trade equities through Interactive Brokers and crypto through Binance from the same conversation.

This is a meaningful engineering investment. Each adapter handles broker-specific authentication flows, order type mappings, error codes, rate limits, and settlement rules. A futures adapter that handles CME's margin requirements is fundamentally different from a crypto adapter that handles Binance's funding rates. ATOM normalizes all of this so the trader and the agent see one consistent interface regardless of what sits underneath.

AI Agent Execution

ATOM exposes 55 MCP tools that cover the full trading lifecycle. AI agents can research, plan, and execute through structured tool calls.

But execution is not unsupervised.

Human Approval

When an agent proposes a trade, ATOM renders a confirmation block. The block shows:

• Instrument, side, quantity, order type
• Dollar risk and percentage of account
• Impact on portfolio heat
• Position sizing method used
• Current G-Score for the strategy

The trade does not execute until the human approves. This is not optional. For live trading, human-in-the-loop confirmation is the default.

Kill Switches and Drawdown Limits

Every workspace has configurable risk boundaries:

• Kill switch. Immediate halt of all trading activity. Triggerable by the user, the agent, or the platform. One command stops everything.
• Drawdown thresholds. Automated position reduction at defined drawdown levels. Hit 5% daily drawdown and position sizing halves. Hit 10% and trading stops.
• Portfolio heat cap. Total open risk cannot exceed the defined ceiling. New trades are rejected if they would breach the limit.
• Correlation monitoring. Real-time detection of correlation spikes that increase effective portfolio risk beyond what individual position limits suggest.

These controls exist because speed without safety is a liability. An AI agent that can execute in milliseconds without risk boundaries is more dangerous than no agent at all. ATOM's risk framework assumes the agent will make mistakes and ensures no single mistake is fatal.

Workspaces

A workspace in ATOM is an isolated trading context. Each workspace has its own broker connections, asset class focus, strategies, risk parameters, and performance history. Nothing leaks between workspaces.

This matters for traders who operate across multiple strategies or accounts. A crypto momentum workspace and an equity swing workspace can run simultaneously with completely different risk profiles, broker connections, and position limits. The agent in each workspace operates within that workspace's rules only.

Workspaces also serve as the organizational unit for teams. Each member sees only the workspaces they have access to. Audit trails are scoped per workspace.

Security

ATOM isolates every agent and every workspace:

• Per-agent encryption. Every agent's data is encrypted with AES-128-CBC (Fernet), keyed to the agent. No agent can access another agent's data.
• Workspace isolation. Each workspace has its own broker connections, market focus, strategies, and risk parameters. Workspaces do not share state.
• Supabase RLS. Row-level security is enabled on all database tables. Queries are scoped to the authenticated user at the database level, not the application level. There is no bypass.

Voice

ATOM supports three tiers of voice interaction:

Tier	Capability	Use Case
Voice Input	Speech-to-text	Dictate queries while watching charts
Voice Chat	Speech-to-text + text-to-speech	Hands-free conversation with ATOM
Live Voice	Realtime duplex audio	Continuous voice interaction, interrupt and redirect naturally

Voice is not a novelty feature. Traders who watch multiple screens need hands-free interaction. Ask ATOM for a risk update, get an audio response, keep your eyes on the order book. Each voice tier is billed per use through compute credits.

Getting Started

1. Sign up at app.systemr.ai. You get $5 in free compute credits. No credit card required.
2. Set up a workspace. Connect a broker, choose your asset class focus, define risk parameters.
3. Start chatting. Ask ATOM to analyze a position, run a backtest, or scan for setups. The domain layer does the work.
4. Graduate to execution. When ready, enable trade execution with human approval. Confirmation blocks keep you in control.

For developers building AI trading agents, the SDK provides programmatic access to the same 55 MCP tools and 187 domain services. Install with pip install systemr-sdk and connect with an API key. The MCP tools guide covers the full tool catalog.

For terminal users, the System R CLI provides the same capabilities in a Claude Code-style interface. Same backend, same tools, same risk engine.

FAQ

How is ATOM different from other AI trading platforms?

Most platforms wrap an LLM around market data and call it AI trading. ATOM is a complete operating system with 187 domain services that encode real trading expertise: G-Score, regime detection, Monte Carlo simulation, multi-factor risk engine, position sizing frameworks. The LLM is just the interface. The domain layer is the product. That layer works with any model.

Do I need programming skills to use ATOM?

No. The chat interface is designed for natural language interaction. Ask questions, analyze strategies, and execute trades through conversation. For developers who want programmatic access, the SDK and CLI are available, but they are not required.

What does ATOM cost?

ATOM uses usage-based compute credits. You pay for the AI compute you actually use, not a flat subscription for access. A $5 free credit on signup is enough to explore the platform, run analyses, and test strategies. Credits can be purchased with USDC, SOL, USDT, PYUSD, or OSR tokens. OSR holders get a 50% bonus on credits.

Can ATOM trade automatically without my approval?

By default, all trade execution requires human approval through confirmation blocks. This is a deliberate design choice. Speed without risk controls is how accounts blow up. As you build confidence and define clear risk boundaries, the autonomy level can be adjusted, but the risk framework always stays active.