"AI-powered" has become a sticker people slap on anything. So it's worth being precise about what the AI in NoVo actually does — and, just as importantly, what it deliberately is not allowed to do.

The four cognitive layers

NoVo uses a large language model for the tasks that genuinely call for synthesis and judgment, running in the background so the trading engine never waits on it. Every AI response is constrained to a strict format, so the model can only return what the system expects — no rambling, no improvisation, no hallucinated trade ideas.

  • Market vision. On a regular rhythm, the AI reads live market structure — trend, momentum, key levels — and produces a directional bias that gates entries.
  • Macro regime. A separate layer classifies the broader environment — risk-on, risk-off, or choppy — to frame every other signal in context.
  • Retail psychology. Another layer models the emotional state of the crowd, flagging when conditions look like a trap and quietly making the system harder to convince.
  • Written market reports. At set points in the session, NoVo writes plain-language tactical breakdowns to your dashboard — structure, levels, and the plan for the next stretch. It's the system thinking out loud, so you're never staring at a black box.

The AI is a co-pilot for judgment, not the pilot. It informs and frames — the mechanical rules, the gates, and your risk limits decide.

A memory that compounds

NoVo keeps two kinds of memory: a fast in-memory layer for the live session and a permanent database for the long term. Everything survives restarts, upgrades, and moving to a new machine. Every trade it has ever taken is recorded with full context — the conditions at entry, how it was managed, how it exited, and the result. Over time that becomes a searchable ledger of every decision the system has made, which is also the raw material it learns from.

Rewriting its own rules

Here's what makes NoVo more than a static program. On a weekly cadence it reviews its own recent outcomes and distills them into concise written rules — durable lessons in the spirit of "in this kind of condition, doing X has been costing us; do Y instead." Those rules are stored and then fed back into the system's future reasoning, so its perspective genuinely evolves based on what actually happened, not on a backtest frozen in time.

Separately, NoVo periodically audits its own historical performance to check whether its signals are still as predictive as they were, and nudges its entry strictness accordingly — tightening up if a setting has been letting weak trades through. It's housekeeping that a human would have to remember to do, running automatically.

Why this matters

A fixed rule-bot is frozen the day it ships; the market moves on and it slowly decays. NoVo is built to do the opposite — to accumulate context, keep a brutally honest record, and revise itself on a schedule. None of this is a promise of profit; markets are uncertain and losses are part of any honest system. It's a promise of improvement discipline: the system reviews itself every week, whether or not you remember to.