APeX AI Agent. An AI trading agent you can watch work.
APeX paper-trades the market's top gainers on live minute data — every session, with every decision on the tape. Run your own lineup of agents, launched by hand or on your daily schedule.
Paper trading / research only. Not financial advice. Not an order-execution system.
About APeX
Real-time market intelligence, not just a trading score.
Most trading tools tell you what happened after the move is over. APeX is built to detect, evaluate, and react to fast-moving market conditions in real time while keeping the decision trail inspectable.
In the June 26 live run, the platform processed more than 4.4 million market messages across 35 actively updating symbols. The websocket stayed connected, message freshness was tracked, and multiple strategy agents continued evaluating opportunities until stop time.
The core value is explainability. Agents ranging from strict to aggressive can be evaluated independently under the same market stream, so risk-profile differences are visible instead of being buried inside a single black-box result.
The page below reads the run in that order: first the audit trail, then the live market stream, then the agent profiles, then the opportunity-capture report that explains what the model caught and what the filters held back.
Monitor real-time market activity at scale.
Evaluate multiple trading styles side by side.
Measure opportunity cost when filters block trades.
Reveal which rules help and which rules need tuning.
Support systematic improvement instead of emotional decision-making.
Audit Trail
How to read a live run.
APeX turns a session into a sequence of evidence: market state, policy behavior, filter reasons, and the tuning record for the next run.
Live market state is recorded first
The run keeps websocket health, message freshness, market phase, aggregate updates, and the active symbol set visible before strategy behavior is interpreted.
Agents are judged under the same conditions
Strict, aggressive, and top-gainer policies evaluate the same stream, so differences in breadth and risk posture can be compared directly.
Every skipped candidate keeps its reason
Symbol classification, spread, and entry gates are reported with the candidates they affected, so the system can explain why names were traded, skipped, or blocked.
Captured and missed moves become evidence
The final review ties market flow, agent behavior, filter pressure, and missed-candidate behavior into a repeatable tuning record.
June 26 log
Real-time flow, kept auditable.
The system processed live trade, quote, and aggregate updates until stop time while tracking connection health, message freshness, and the active symbol set.
Agent profiles
Risk profile changes the tradable universe.
Active agents are compared under the same market conditions. Zero-activity entries are omitted so the table stays focused on meaningful differences.
Symbol Breadth
Risk lanes appear in the block counts.
| Agent | Risk profile | Symbol blocks | Avg return after block | Max return after block | Read |
|---|---|---|---|---|---|
| aggressive_plus | Risk-on plus | 7 | 52% | 239% | Wider appetite, but symbol-class gates still removed runners. |
| aggressive_safe | Guarded risk-on | 3 | 94% | 239% | Fewer blocks, but the missed names were strong. |
| aggressive_early | Early entry | 4 | 81% | 239% | Early policy still narrowed the universe by class. |
| top_gainer_capture | Top gainer | 10 | 68% | 239% | Most visibly constrained by the symbol-class gate. |
| top_gainer_capture_plus | Top gainer plus | 11 | 64% | 239% | Broadest capture variant, but also the most filtered. |
Aggressive variants blocked 3 to 7 candidates, which shows narrower universe selection even inside risk-on policies.
Top-gainer policies blocked 10 and 11 candidates, so their breadth was shaped most visibly by the symbol-class gate.
The blocked names later showed large returns, which is useful attribution for tuning risk gates rather than a P/L claim.
Opportunity Capture
Filters reveal the opportunity cost.
The aggressive_safe report separates the move the market offered, what the model captured, and which filters held candidates back.
Why missed runners matter
The audit shows both capture and restraint.
In the aggressive_safe report, APeX completed six trades with no open positions and an 11.68 profit factor. SDOT was the standout entry: the model traded it during a 238.67% candidate move and captured 46.55% of the favorable excursion.
The same report recorded 14 blocked candidates: 11 by spread filters and 3 by entry filters. That separates what the market offered from what the agent actually acted on.
Top missed candidates
blocked_bySpread filters carried most of the blocks
The aggressive_safe opportunity report recorded 11 spread-filter blocks, including PCLA, IVF, WSHP, ILLR, and RYOJ.
Entry filters flagged overheated flow
Three candidates were blocked by entry filters, including SKYQ at 27.08%. The report keeps those misses attached to their rule reason.
Missed runners create the next experiment
The scanner basket return was 47.67%, with missed candidates including PCLA at 125.73% and IVF at 102.42%. Those are tuning signals, not performance promises.
APeX AI Agent is experimental paper-trading research software. It does not place orders with any broker, and status figures are point-in-time examples — not an offer, recommendation, guarantee, or investment advice. Trading involves risk, including the possible loss of principal.