Skip to main content

Risk And Security

HyperionX is trading software. Treat every feature that can affect orders, accounts, wallets, or strategy execution as an account-risk boundary.

Operating Rules​

Use these rules in every account-affecting workflow:

  • Begin in LocalPaper and verify the selected account before each action.
  • Treat AI as read-only unless a documented action is deliberately enabled.
  • Do not assume a confirmation dialog exists; Chart Trader and MegaDOM can submit directly.
  • Keep credentials and private keys out of scripts, logs, prompts, profiles, and AI context.
  • Keep the Preview local API on loopback with a private desktop token.
  • Preserve action audit records, but verify fills and orders at the authoritative provider.
  • Keep research results separate from claims about live performance.

Trading Risk​

Trading systems can fail for many reasons:

  • Bad strategy logic.
  • Poor historical data.
  • Overfitting.
  • Provider outages.
  • Slippage.
  • Fees.
  • Latency.
  • Partial fills.
  • Incorrect tick size or lot size.
  • Unexpected order state mapping.
  • Connection disconnects.
  • User configuration mistakes.

Backtests and optimizations are research tools. They do not guarantee live results.

AI Risk​

AI can help users move faster, but it can also produce incorrect code, wrong assumptions, or overconfident analysis.

AI-generated output should be treated as a draft until verified. Cloud providers can receive the prompt and chart, strategy, account-label, research, or script context supplied for the request. Ollama remains local only when its configured endpoint is local.

Safe AI rules:

  • Compile AI-generated scripts before using them.
  • Review strategy logic before enabling.
  • Run in simulation before live use.
  • Keep AI read-only unless a tool is explicitly enabled.
  • Keep secrets out of prompts, custom parameters, profile files, and feedback instead of relying only on automatic redaction.
  • Begin Local Desktop API trading on Hyperion Sim. Use release-gated Hyperliquid or KuCoin only when the matching provider capability and setting, entitlement, active account/instrument, connection, and exchange-permission checks explicitly enable it.
  • Keep audit logs of agent actions.

The managed Rion cloud desktop bridge is an Internal HyperionX component. It is not a public integration surface. See AI Data and Privacy for provider destinations, local storage, and retention.

Credential Safety​

The current chart-context parameter filter omits names containing key, secret, password, or token. Other AI feedback paths use additional pattern-based redaction. These filters do not reliably classify every user-defined name or value.

Never put API keys, passwords, bearer tokens, private keys, seed phrases, mnemonic phrases, or wallet material in:

  • Code Lab parameters or source
  • AI prompts and chat messages
  • Agent Builder instructions, prompt files, or skill files
  • local API request fields that do not explicitly require the value
  • logs or public support exports

Connection and AI settings use protected local storage where implemented on Windows, but profile files, application databases, audit logs, and support bundles can still contain sensitive metadata. Review them before sharing.

Wallet Safety​

Never paste a wallet private key into Code Lab source, strategy parameters, AI prompts, log files, or plain-text configuration.

Recommended wallet model:

  • Use an external wallet or secure wallet bridge.
  • Build transactions inside HyperionX only after explicit user action.
  • Sign through the wallet.
  • Show clear transaction details before signing.
  • Keep AI read-only until transaction tools are explicitly permissioned.

Local API Safety​

The Local Desktop API is Preview. Current safety boundaries are:

  • Bind only to 127.0.0.1.
  • Try ports 5217 through 5226.
  • Require the desktop token for every request except a CORS preflight, including health requests.
  • Use a preconfigured token for an external integration; there is no supported public UI that reveals the generated process token.
  • Check route permissions, license entitlements, and local settings independently of authentication.
  • Treat Agent Builder profile controls and HTTP-token permissions as separate boundaries.
  • Keep direct trading actions on Hyperion Sim. A caller-provided confirmation flag is not an independent desktop approval dialog.
  • Include build/version metadata.
  • Audit action attempts. Request bodies are replaced with a redacted length marker rather than stored in the audit file.
  • Do not expose the listener or token to a LAN, internet service, browser page, or untrusted local process.
  • Do not depend on the Preview interface without checking the installed build's capabilities.

See Local Desktop API for the routes, token-provisioning limitation, permissions, and current unavailable features.

Strategy Safety Checklist​

Before trusting a strategy:

  1. Compile without errors.
  2. Confirm it uses HyperionX APIs.
  3. Guard early bars.
  4. Validate indicator values.
  5. Test on historical data.
  6. Test out-of-sample.
  7. Include supported fees and account separately for slippage; the current research simulator has no configurable slippage control.
  8. Run in playback or simulation with an explicitly selected LocalPaper account. Playback does not select it automatically; the current release-candidate build rejects external-account order mutations while Playback is active.
  9. Confirm order sizes, stops, and targets.
  10. Monitor live behavior with small size first.

UI Safety​

Color, disabled styling, a selected OCO preset, or a visible account label is not an independent enforcement control. Before an action, read the full account, instrument, Qty/USD mode, size, order type, price, and protection. Confirm the result in Orders, Positions, and Executions and at the external provider when applicable.

Support Logs​

Support logs are useful for diagnosing provider, script, and workspace problems. Best-effort redaction is not a guarantee, so inspect every export before sharing it.

Useful log categories:

  • Connection status.
  • Provider errors.
  • Code Lab compile diagnostics.
  • Indicator runtime exceptions.
  • Strategy runtime exceptions.
  • Workspace restore errors.
  • Order state transitions.
  • AI provider errors without secret values.