Define the rules
Start with a persona, watchlist, scan cadence, risk limits, and trading thesis. The agent needs a frame before it can be reviewed.
Use case
AI paper trading agent
Create a paper-trading agent around your rules, run it in simulation, and review every decision with transparent context before any live-capital workflow is considered.
Run in paper mode
The agent evaluates market context and records simulated decisions. Alerts are workflow notifications, not buy or sell signals.
Review the process
Use the journal, decisions, risk checks, and benchmark context to see whether the rules are behaving the way you expected.
| Workflow step | Trading Boy surface | Review question |
|---|---|---|
| Choose a frame | Persona library | Does this agent match the way you want to practice trading? |
| Create the agent | CLI and hosted API | Are watchlist, cadence, and guardrails explicit? |
| Track decisions | Journal and decision history | Can you explain why the agent acted? |
| Review risk | Risk controls and leaderboard context | Do drawdown, sizing, and frequency stay inside the intended process? |
Important boundary
Trading Boy is paper-trading software. It does not execute live trades, hold funds, or provide financial advice.
How to use an AI paper trading agent safely
The useful output of an AI paper trading agent is not a prediction that should be copied into a live account. The useful output is a decision trail. A good paper agent makes its assumptions explicit, records what it saw, shows how the rule fired, and gives you enough context to decide whether the workflow deserves more testing.
Start by writing the agent as a narrow practice system. Define the market universe, the setup it should look for, the time horizon, the maximum paper position size, and the condition that invalidates the trade. If those rules are vague, the review will be vague too. Trading Boy works best when the agent is treated as a structured paper-trading workflow rather than a general chat prompt that comments on price action.
After the agent records a simulated entry, use the journal and risk review pages to inspect the decision. Ask whether the agent respected the rule, whether the entry matched the intended setup, whether the stop or invalidation was clear, and whether the position size would still make sense if the trade moved against the thesis. The goal is to find repeatable process evidence, not to celebrate or discard the workflow after one result.
When a paper agent creates repeated mistakes, change one variable at a time. Tighten the setup definition, reduce frequency, adjust the watchlist, or add a risk filter, then keep the next sample separate from the old one. That gives you a clean comparison between versions and makes the paper-trading record more useful for future review.
Example paper-agent review
Scenario: A momentum paper agent watches a small crypto watchlist and logs a simulated long entry after a breakout. The agent records the setup, the entry price, the invalidation level, and the reason it believed momentum was still present.
Review: The trader checks the pre-trade checklist and sees that the invalidation was clear, but the position size was higher than the written paper risk limit. The result of the trade is less important than the process error: the agent had enough setup context, but the risk rule was not enforced tightly enough.
Next action: The trader does not promote the workflow or rewrite the entire persona. They add a hard sizing rule, run the paper agent for another sample, and use the post-trade review page to compare whether the risk breach repeats.
Signals to trust less
Be careful when a paper agent performs well over a tiny sample, only works in one fast market, ignores drawdown, or cannot explain why a decision matched the written rule. Those are review prompts, not proof that the workflow is ready for live capital.
Signals to keep studying
A stronger workflow records clear reasons, follows the same constraints across wins and losses, avoids oversized paper positions, and creates journal entries that a human can audit later without guessing what happened.
AI paper trading agent cluster
Use these supporting pages when the agent idea needs more structure. The pillar page explains the product use case; the cluster pages help write the rules, turn them into prompt instructions, control paper risk, and evaluate the sample without relying on paper profit alone.
| Cluster page | Use it when | Output |
|---|---|---|
| AI trading agent rules | The agent idea is still vague. | A versioned rule set with setup, invalidation, skip, and review fields. |
| AI trading agent prompt template | The rules need to become agent instructions. | A paper-first prompt with risk boundaries and journal output. |
| AI trading agent output format template | The agent output varies too much to review. | A consistent paper-decision record with required fields and skip reasons. |
| AI trading agent prompt versioning | A rule or prompt change needs a clean audit trail. | A version log that separates prompt drift from strategy performance. |
| AI trading agent without live execution | The agent workflow needs a clear simulated-only boundary. | A paper-mode review process without custody, live orders, or signal framing. |
| Simulated AI trading agent review process | A sample of agent decisions is ready for review. | A repeatable review sequence for entries, skips, exits, and caveats. |
| AI agent risk controls | The agent needs hard paper limits. | Size, drawdown, exposure, invalidation, and pause controls. |
| AI paper trading agent workflow | The agent idea needs a full setup-to-review operating loop. | A staged workflow for rules, prompts, output, risk, decision review, and versioning. |
| Paper trading agent risk control workflow | The sample needs size, exposure, drawdown, skip, and pause checks. | A risk-control review that treats paper wins and losses with the same rules. |
| AI trading agent decision review workflow | Entries, skips, and exits need row-by-row review. | A decision review record before prompt or rule changes. |
| AI trading agent human review checklist | A reviewer needs to approve the next paper-mode action. | A human checklist for sample quality, risk behavior, evidence, and next action. |
| AI paper trading agent evaluation | A paper sample is ready to review. | A scorecard for rule fit, risk behavior, journal quality, and version changes. |
| AI paper trading agent evaluation checklist | The evaluation needs a yes/no quality gate. | A checklist for evidence completeness, risk discipline, sample quality, and next action. |
AI paper trading agent FAQ
What is an AI paper trading agent?
It is a simulated trading workflow that applies written rules, logs paper decisions, and creates review evidence without placing live trades or controlling funds.
Does Trading Boy execute live trades?
No. Trading Boy focuses on paper trading, decision review, journaling, risk controls, and educational workflow testing.
What should I review before changing an agent rule?
Review the thesis, invalidation, sizing, drawdown, frequency, sample size, and whether the decision followed the rule. Change one variable at a time so the next paper sample is meaningful.