The pre-trade risk gate
for AI trading agents.
One call between your agent and the broker. Guard answers, in milliseconds: should this exact order happen right now? The decision comes back signed, bound to the order, and dead in 60 seconds.
What one decision looks like
POST /v1/trade/preflight
{
"order": { "symbol": "SPY", "side": "buy",
"notional_usd": 5000,
"order_type": "limit_nbbo" },
"policy": { "max_notional_usd": 10000,
"max_cost_bps": 25,
"daily_loss_limit_usd": 200,
"confirm_above_usd": 8000 },
"state": { "daily_pnl_usd": -40,
"open_positions": 1 },
"evidence": { "audit_score": 95 }
}{
"decision": "REDUCE",
"approved_notional_usd": 2500,
"reasons": [{ "code": "VOLATILITY_ELEVATED",
"message": "VOLATILE regime - position
scalar 0.5 applied" }],
"order_hash": "sha256:b971f9ef…",
"expires_at": "…60 seconds from now",
"attestation": {
"signer": "0x26e3F3e2…",
"signature": "0x…"
}
}What goes into every decision
Your policy, enforced deterministically
Max notional, max cost in bps, daily loss limit, symbol allowlist, open-position cap, human-confirmation threshold, minimum audit score. Same inputs, same answer, every time. The LLM explains; it doesn't decide.
Live volatility regime
A validated variance detector over 417k+ snapshots: VOLATILE predicts roughly 2x forward volatility with ~90% persistence. In a VOLATILE tape, approved size scales down automatically. Stale feed? Guard fails closed.
Execution-cost reality
Spread, slippage and square-root size impact, anchored to costs we measured on live fills (1.5bps round trip on liquid ETFs). If estimated cost eats your policy ceiling or most of your claimed edge, the trade doesn't pass quietly.
Strategy evidence
Feed in the strategy's Reality Check audit score. Failed forensics means BLOCK. No evidence means the decision requires human confirmation. Verified evidence unlocks autonomy. That's the ladder.
The attestation is the product
Broker order review checks the transaction. Guard certifies the decision. Every response is signed with EIP-191 by a published key (pin it here), bound to one exact order by hash, and expires in 60 seconds. An agent can't get approval for 10 shares and submit 100. A platform can log, replay and prove every decision its agents made. That audit trail is what compliance teams, prop firms and brokers actually need from agentic trading.
Integrate in an afternoon
- REST:
POST api.sygnliq.com/v1/trade/preflight- OpenAPI docs - MCP: native server at
api.sygnliq.com/mcp, tooltrade_preflight- listed in the official MCP registry ascom.sygnliq/sygnldata - x402: agents pay $0.01 per decision in USDC on Base, no signup. Free rate-limited demo at
/v1/trade/preflight/demo - Enterprise: contracts, SLAs and design-partner pilots for brokers, prop firms and agent platforms: [email protected]
FAQ
What is a pre-trade risk gate?
An independent check between an AI agent's trading decision and the broker's order endpoint. Guard evaluates the proposed order against your policy, current volatility, realistic costs and the strategy's audit evidence, then returns a signed, expiring ALLOW / REDUCE / REQUIRE_CONFIRMATION / BLOCK decision.
Does it work with Robinhood agentic trading?
Guard is broker-neutral. It runs in the agent's workflow before any order is submitted, so it works the same with Robinhood's Agentic Trading, Tradier's MCP server, or your own brokerage integration. Guard never touches your brokerage credentials.
Is this investment advice?
No. Guard deterministically evaluates your policy against measured market state and modeled execution cost. It never picks symbols, predicts direction, or promises returns.