Chapter I
What is AI trading?
An encyclopaedic overview: LLM agents, machine-learning signals, classical quant, and where each fits in the retail-to-institutional spectrum in 2026.
Read the chapter →Contents
Chapter II
LLM agents vs quantitative algorithms
Where large language models add real information, where they hallucinate, and how they interact with traditional statistical strategies.
Read the chapter →Chapter III
Broker & API setup
Interactive Brokers, Alpaca, Schwab, Trading 212, Saxo, eToro, Public, Tastytrade — retail-accessible APIs, auth patterns, paper trading and 2026 status.
Read the chapter →Chapter IV
Building the Python stack
Backtesting frameworks, market-data providers, feature libraries, LLM-agent orchestration — 2026 status with pricing and maintenance detail.
Read the chapter →Chapter V
Multi-agent architecture
Separation of powers between research, portfolio, risk and execution agents — the pattern this reference recommends.
Read the chapter →Chapter VI
Risk & governance
Deterministic risk engines, kill-switch conditions, drawdown controls and the model-oversight layer.
Read the chapter →Chapter VII
Regulatory boundary
UK FCA (MAR 7A, PERG 13, Consumer Duty), US SEC/FINRA/CFTC and EU MiFID II + AI Act. Where personal use ends and regulated activity begins.
Read the chapter →Reference
Glossary & sources
Working definitions of the vocabulary used across the site, plus a full source list — primary regulatory documents, docs, papers.
Read the reference →Editorial position
The safest and most useful design is not a single autonomous bot.
This reference takes a clear editorial view. A single autonomous trading agent — one language model receiving news, deciding, sizing, executing — is the wrong shape. It is difficult to audit, easy to hallucinate, impossible to fully constrain, and it puts broker credentials directly in the hands of a system whose failure modes include producing persuasive-sounding but wrong reasoning.
The design pattern this site recommends is a controlled team. Specialist agents research and challenge ideas. Deterministic software calculates portfolio limits, position sizing and order validation. An independent risk engine can reduce or veto any proposed order. A restricted execution service — not an open-ended language model — is the only component with broker access. Every decision is auditable end-to-end.
The full architecture, including the eleven agent roles and the controlled decision workflow, is set out in a companion blueprint at agentteam.co.uk/share-trading.html. That page is the authoritative statement of the pattern. This site sits alongside it as an extended technical reference: how to actually build and run the components.
Companion blueprint
The eleven-agent blueprint, at agentteam.co.uk
The full separation-of-powers design — agent roles, controlled decision workflow, formal trade proposal schema, kill-switch conditions, seven-stage deployment roadmap. Read alongside the technical chapters here.
Read the blueprint →