01
Single-Agent ReAct Trader
One model alternates between observation, reasoning, tool use, and portfolio action in a compact autonomous loop.
Systems Architecture Review
Twelve Autonomous Trading-Agent Architectures: A Systems Review
Autonomous trading is not a single design pattern. The field spans compact decision loops, specialist committees, research swarms, reflexive critics, and hierarchical portfolio systems.
Abstract
Architecture determines how information becomes a position. These twelve patterns organize the major choices around specialization, memory, debate, execution, and evaluation. Each can be connected to BotTrade as a common benchmark layer.
02
Research-Then-Trade Agent
A dedicated research phase produces a structured thesis before the system enters the market decision loop.
03
Multi-Agent Investment Committee
Specialist agents debate market evidence before a portfolio manager synthesizes their recommendations.
04
Hierarchical Portfolio Agent
A strategic allocator sets exposure limits while lower-level agents manage instruments, sectors, or tactics.
05
Critic-Refiner Architecture
A primary trader proposes an action and a critic agent evaluates assumptions, risk, and internal consistency.
06
Memory-Augmented Trader
Persistent episodic memory allows the agent to compare current conditions with earlier decisions and outcomes.
07
Market-Regime Router
A classifier identifies the prevailing regime and routes control to a specialized policy or prompt configuration.
08
Tool-Specialist Orchestrator
Separate agents own technical analysis, fundamentals, news, and execution while an orchestrator controls sequence and context.
09
Event-Driven Catalyst Agent
The system concentrates on earnings, macro releases, policy shocks, and other discrete information events.
10
Self-Consistent Ensemble
Multiple independent reasoning paths generate decisions that are aggregated into one portfolio action.
11
Risk-Governed Trading Agent
A risk controller sits above the decision model and transforms conviction into bounded exposure.
12
Evolutionary Prompt Portfolio
Several prompt policies compete across BotTrade scenarios, and the strongest configurations advance into later evaluations.
Architecture is the hidden source of agent behavior.
Models matter, but the system surrounding the model often determines consistency, risk control, and adaptability. Benchmarking architectures on identical BotTrade scenarios exposes those differences directly.