Skip to content
CTAgent

Guardrails & Security

CTAgent is built on one principle: the AI proposes, deterministic code disposes. The agent can design and suggest trades, but every real order is validated and executed by plain, audited code with safety rules the model cannot bypass.

Trading guardrails

Every live order must pass through a deterministic trade-execution layer that enforces:

  • Symbol allow-list — strategies can only trade approved pairs. Nothing else gets through.
  • Max order size — a hard cap on the size of any single order.
  • Daily-loss limit — a per-user, ledger-based limit. Hit it, and trading stops for the day.
  • Kill-switch — a global and per-user emergency stop (/killswitch) that halts all trading instantly.

These checks are fail-closed: if anything is uncertain, the order is rejected rather than risked.

Your keys are never exposed to the AI

  • API keys are encrypted at rest in a secrets vault.
  • They are loaded only by the deterministic execution code that talks to Binance — never passed to the AI model, never logged, never echoed back in chat.
  • Keys must be trade-only (withdrawals disabled) and IP-restricted. CTAgent is non-custodial; your funds never leave your Binance account. See Connect Binance.

Strategies are declarative, not code

CTAgent strategies are structured rules — indicators, thresholds, sizes — not arbitrary code. The AI never writes or runs executable trading code. That removes an entire class of risk: there’s no code-execution surface for a strategy to abuse.

Spot-only, paper-first

  • Spot only — no futures, margin, or leverage. You can’t be liquidated.
  • Paper-first — every strategy starts in simulation. Live trading is a separate, explicit, double opt-in.