AgentNash
What is AgentNash?How It WorksConnecting Your AccountTraining vs Live ModeStrategiesCouncil V2The SuperforecasterThe Council (Legacy)TerminalAgentsBenchmarkingSafeguards & RulesNuclear Option

System

Safeguards & Rules

Two-tier rules engine protecting every trade on every exchange.

Every trade passes through a two-tier rules engine before execution. No order can reach any exchange (Kalshi or Polymarket) without passing both tiers. The system is deterministic — no AI judgment in the validation layer, only hard programmatic checks.

Tier 1 — Per-Agent + Global Rules

Nine constraints are checked against both the agent's per-agent config and the account's global rules. When both define a limit, the more restrictive value wins.

RuleDefaultDescription
Max trade size$100Rejects trades exceeding this cost
Capital per agent$2,000Max capital a single agent can use
Daily loss limit$500Pauses agent if daily losses exceed this
Min confidence60%Rejects trades below this confidence score
Allowed categoriesAllWhitelist of tradeable market categories
Blocked tickersNoneBlacklist of specific market tickers
Max positions10Concurrent open position limit
Duplicate preventionOnBlocks same agent from opening duplicate position
Opposing positionBlockedPrevents YES and NO on same market

Tier 2 — Account-Level Validation

Account-wide checks that apply across all agents. These are tracked per-environment — training and live limits are counted independently.

RuleDefaultDescription
Max trades / day50Daily cap across all agents combined
Global daily loss$500Pauses all trading if total losses exceed this
Max trades per marketUnlimitedLimits total trades on any single market
Cooldown0 hoursMin time between same-market trades
Active hoursAlwaysRestrict trading to specific UTC hours
Daily AI budget$50Max daily LLM API spend across all agents

Configuring Rules

All rules are configurable from Settings → Safeguards. Changes auto-save automatically and take effect on the next validation cycle — no restart or redeployment needed.

How Conflicts Resolve

  • Size limits — MIN(agent value, global value). The lower limit wins.
  • Confidence floors — MAX(agent value, global value). The higher requirement wins.
  • Category filters — Agent categories OR global categories apply.
  • Blocked tickers — Global only (not configurable per-agent).

Rejected trades are visible in your trade history with the exact rule that blocked them and an explanation. Use this to tune your settings.