Technical documentation

How AQ Agent works, under the hood

This is the technical side of the product, separated from the marketing landing. Here live the architecture, the rules behind each engine, the risk controls, and how to reproduce the backtests. Order execution and risk management are fully deterministic and rule-based. The one AI-assisted step is the Active Select engine's monthly stock list — proposed by an AI model, frozen with its rationale for audit, and validated forward in paper trading before it can earn more weight.

Architecture

Each client runs their own agent on their own brokerage account — never a shared pool. The system has three pieces:

  • The agent runner evaluates each engine's signals every cycle (roughly every market hour) and submits orders to the broker.
  • The scheduler fires the cycles and reconciles capital against your portfolio's target.
  • The dashboard shows live positions, P&L, decisions and costs — read-only over the broker's state.

Broker credentials are stored encrypted (Vault), scoped to trading only. The agent has no fund-withdrawal connection.

Engines & signals

AQ Agent groups rule-based strategies into five engines. Your portfolio decides which are available; you choose which run.

Market Core — index-based equity allocation
Holds SPY (or a SPY + QQQ split in the higher portfolios) while the S&P 500 sits above its 200-day moving average. When the index drops below, it rotates 100% into 1-3 month T-bills (BIL) and waits. The primary return engine.
Active Select — monthly model-assisted selection
A concentrated basket of individual growth stocks, from AQ Plus up. Every month an AI model re-reads a universe of liquid US large- and mid-caps and rebuilds the shortlist this engine may trade — the same list for every subscriber who enables it, frozen and fingerprinted before use. A deterministic 9-signal engine manages entries/exits daily under a crash guard and correlation pruning. By design it carries no backtested return claim — an AI-assisted selection cannot be honestly backtested — so every list is validated forward in paper against the S&P 500.
Adaptive Defense — multi-trend defensive allocation
Holds each asset class (US and international equities, bonds, REITs, commodities, gold) only while its long-term trend is positive; rotates to cash when it breaks. Built to sidestep prolonged bear markets.
Range Income & Covered Income — defined-risk iron condor, automated wheel
Sells option premium with bounded risk. The iron condor — where the cost of the index leaving its range is fixed at entry — runs from AQ Pro up. Most of its capital sits in a short-term US Treasury ETF at all times; it commits a smaller portion to spreads only when option premiums are high compared with their own past year. The wheel (cash-secured puts that roll into covered calls) is AQ Studio only: each contract ties up the full purchase amount, which is what sets that plan's funding figure. These trades pay minimal regulatory/clearing fees per contract.
Crypto — rules-based crypto allocation
A BTC + ETH allocation gated by a 100-day moving average, volatility-targeted so it never dominates portfolio risk. Steps aside (cash) when the trend breaks. AQ Studio only, and off until you switch it on.
Custom Portfolio — positions you name yourself
The one engine where you choose the stock: you name the symbol, the size and the entry, optionally with a target and a stop. The agent then runs it under the same controls as every other position — volatility-based stops, trailing protection, size caps and the portfolio circuit breaker. Available on AQ Pro and AQ Studio.

Portfolio allocation

Each portfolio blends the engines above at fixed target weights. Every level adds engines on top of the one below; within your portfolio you turn the available ones on or off from Settings. Tap a portfolio to see its mix.

  • Market Core — Index-based equity allocation 70%
  • Adaptive Defense — Multi-trend defensive allocation 25%
  • Reserve — Short-term Treasury-bill reserve 5%
  • Market Core — Index-based equity allocation 35%
  • Active Select — Monthly model-assisted selection 35%
  • Adaptive Defense — Multi-trend defensive allocation 25%
  • Reserve — Short-term Treasury-bill reserve 5%
  • Market Core — Index-based equity allocation — SPY + QQQ 30%
  • Active Select — Monthly model-assisted selection 30%
  • Adaptive Defense — Multi-trend defensive allocation 20%
  • Custom Portfolio — User-selected portfolio 10%
  • Range Income — Defined-risk Iron Condor strategies 5%
  • Reserve — Short-term Treasury-bill reserve 5%
  • Active Select — Monthly model-assisted selection 25%
  • Covered Income — Automated Wheel strategy 20%
  • Market Core — Index-based equity allocation — SPY + QQQ 15%
  • Crypto — Rules-based crypto allocation 10%
  • Custom Portfolio — User-selected portfolio 10%
  • Adaptive Defense — Multi-trend defensive allocation 10%
  • Range Income — Defined-risk Iron Condor strategies 5%
  • Reserve — Short-term Treasury-bill reserve 5%

Every included engine can open a position from $400 / $400 / $500 / $1k by plan — the buying power the engines mechanically need, not a suggested account size. Opt-in engines (options, crypto) state their own requirement when enabled; below any floor the rest still run.

Risk controls

  • Per-position loss limit: every position is sized against a loss level set before it opens (wing width on spreads, ATR stop on equity). A stop is an instruction to exit, not a guarantee of the exit price — a gap can fill it lower.
  • Drawdown circuit breaker: if portfolio drawdown crosses your portfolio's budget, the agent flattens everything.
  • Volatility target: the crypto engine is sized so it can't dominate the book's risk.
  • Regime filter: the growth and defensive engines step into cash when their long-term moving average turns over.
  • Paper-first gate: the Active Select engine's AI-selected lists must build a positive live paper track record before their allocation can grow.

Backtests & reproducibility

Per-strategy backtests (wheel, iron condor, cash-secured put, multi-asset trend, crypto momentum) and per-portfolio compositions are versioned in the product repository, with their methodology and caveats:

  • services/core/backtests/ — per-strategy backtests
  • docs/validation/bundles/ — per-portfolio compositions

Backtest results are hypothetical and forward-looking, not realized returns. They are calibrated against SPY's 2015-2024 baseline and assume the 200-day regime filter behaves consistently with that calibration. Higher growth potential means higher volatility. Past performance does not guarantee future results.

Custody & security

  • Your money never leaves your broker. All trading happens in your own account.
  • API with no withdrawal access. Broker keys are trade-only; they cannot move or withdraw funds.
  • Encrypted credentials. Stored in Vault, rotatable from the dashboard.
  • Paper-first. 60 days of simulation before you can activate real capital.

Source code

The engine, the strategies and the backtests live in the product repository. If you'd like to audit the methodology or run the backtests yourself, reach us at hello@alphaquantagent.com.

AQ Agent is a software service, not a registered investment adviser, broker-dealer, or fund manager. All trading happens in client-owned brokerage accounts. Historical backtest results are hypothetical and do not guarantee future results. Investing carries risk, including possible loss of principal. AQ Agent is not affiliated with Alpaca Securities LLC.