Skip to main content

Brain

The brain is the agent’s model plus its trader personality — raw intelligence and how it carries itself into a trade. It is written into the prompt the model reasons with. It never edits your strategy modules or protocol; those stay on the strategy.

Defaults on a blank draft come from platform config. Today that usually means Claude Haiku 4.5 and Moderate · Realist · Measured.

Quick map​

PieceWhat it does
ModelWhich approved LLM decides on each signal
RiskHow hard it is willing to stake
OutlookBias when signals are ambiguous
ConvictionHow much confirmation it needs before committing

Model​

Only approved models appear in the picker (grouped by provider). Each row shows three meters from Artificial Analysis where published — scores are 1–5, or blank when a metric is not available:

MeterMeasuresHigher meansLower means
IntelligenceReasoning depth on conflicting evidenceCatches subtle, multi-signal setups; fewer false entries on borderline tradesDecides quickly on clear patterns; can miss nuance when signals disagree
CostSpend from your daily LLM credit pool per decisionPremium pricing — fewer evaluations per day on the same creditsCheap to run — more headroom to evaluate signals
SpeedHow fast a decision returns once a signal triggersSnappy — useful when entry timing is tightSlower — fine for slower cadences where seconds matter less

Pick the mix that fits the job, not the biggest name:

  • Fast / frequent — lower Cost, higher Speed; Intelligence can stay modest.
  • Hourly / swing — balanced across the three.
  • High-conviction / messy signals — raise Intelligence; Cost and Speed matter less.

If a previously saved model is no longer approved, the UI shows Model unavailable — pick another before you can save.

Reading the live catalog​

The picker is the source of truth (models change). Snapshot of how to read a few common picks:

ModelTypical read of the meters
Claude Haiku 4.5 (catalog default)Modest Intelligence, mid Cost, strong Speed — solid all-rounder for frequent evaluation
Claude Opus / GPT flagship tiersHigher Intelligence and Cost — fewer, deeper reads
Flash / Nano / low-Cost rowsCheap evaluation budget — more signals per credit day
Pinned / provider-featured rowsSame meters; still choose by Intelligence · Cost · Speed, not the pin

Meters are advisory for selection. They do not change your Risk management rails.


Personality (trader soul)​

Three axes. Each choice has a short label in the UI and a matching instruction the agent is told to follow when it reasons.

Risk — how big you stake​

Appetite for position size and variance.

ChoiceIn plain wordsWhat the agent is told
ConservativeModest commitment; fewer, higher-confidence bets; steadier outcomesStake cautiously — keep commitment modest and prefer fewer, higher-confidence bets. Accept lower variance for steadier outcomes.
Moderate (common default)Commit when a setup earns it; don’t overextend on one thesisStake with balance — commit meaningfully when a setup earns it, but never overextend on a single thesis. Tolerate moderate variance for a solid result.
AggressiveConcentrate on strongest reads; chase larger movesStake boldly — concentrate behind your strongest reads and chase the larger move. Accept higher variance and drawdown for an outsized result.

Risk personality is not Max Exposure or Small/Medium/Large %. Those are hard rails on the Risk Management step. Personality biases how the model uses the size presets it is allowed.

Outlook — which way you lean​

How ambiguous signals are read, and the confidence baseline.

ChoiceIn plain wordsWhat the agent is told
OptimistAmbiguity → opportunity; warmer confidence; breakouts get the benefit of the doubtRead ambiguous signals as opportunity and lean toward upside continuation. Start from a warmer confidence baseline and give breakouts the benefit of the doubt.
Realist (common default)Face value; no directional lean; confidence from evidence onlyRead signals at face value with no directional lean, weighing opportunity and trap evenly. Anchor confidence strictly to the evidence in front of you.
PessimistAmbiguity → trap; guarded confidence; setups must prove themselvesRead ambiguous signals as a trap and stay alert to downside reversals. Start from a guarded confidence baseline and make setups prove themselves first.

Conviction — how much evidence before committing​

ChoiceIn plain wordsWhat the agent is told
CautiousHigh bar — several independent signal categories must agreeRaise the evidence bar — require confirmation from several independent signal categories before committing. Prefer setups where multiple categories agree.
Measured (common default)Primary signal + one confirming read from another categoryHold a balanced evidence bar — a primary signal plus one confirming signal from a different category is enough. Weigh confirmation and opportunity evenly.
BoldLower bar — strong single-category conviction can leadLower the evidence bar — act on strong single-category conviction. A powerful momentum signal or clean structure breakout can suffice without full confirmation.

How Brain pairs with Strategy and Risk​

  1. Strategy decides what the agent sees (modules, rules, timeframe).
  2. Brain decides how the model interprets that evidence (model + soul instructions).
  3. Risk management decides whether a sized trade is allowed (exposure, stops, sizing %).

Same playbook + Aggressive · Bold · Optimist will press harder and earlier than Conservative · Cautious · Pessimist — still inside the same capital rails.

Scenario sketches​

IntentBrain sketch
Steady Radar autopilot, limited creditsHaiku or other mid/low Cost · Moderate · Realist · Measured
Catch early breakouts, accept noiseHigher Speed or Bold · Optimist · keep Max Exposure tight
Only take A+ multi-signal setupsHigher Intelligence · Cautious · Realist or Pessimist
Press size when conviction is highAggressive personality and review Large % / Max Exposure on Risk Management — personality alone does not raise the rails

Create vs edit​

You set Brain when you commission the agent (Agents overview). You can change model and personality later on the agent console — the same three axes and approved-model list.

← Overview · Next: Risk management · Hub · Profile