Verified

Trust Checks

Five deterministic pre-flight guardrails an AI agent buys per call before it commits an action or output: schema/type validation, money math, spend-policy enforcement, factual grounding and brand compliance. Each returns a clean pass/fail with reasons.

Trust Checks gives an autonomous agent a cheap, stateless way to verify itself before it acts. Each of the five endpoints takes the agent's proposed output plus the constraints to check it against, and returns a structured pass/fail it can branch on: /schema confirms a structured output has the required fields and correct types; /calculate independently re-computes order/invoice math (gross, fees, net) so totals can't drift; /policy checks an intended action — vendor, amount, category — against allow/block lists and a spend cap before money moves; /knowledge checks an answer against a list of approved facts and forbidden claims to catch ungrounded or off-limits statements; /brand checks copy against forbidden words, required terms and preferred-term substitutions. All endpoints are deterministic, stateless POSTs metered per call via x402 — the caller authenticates by payment (X-Payment header) and needs no API key of its own.

Base URL

https://gateway.apiosk.com/trust-checks

Endpoints

POST /brand $0.07

Check copy against brand rules: forbidden words/phrases, required terms that must appear, and preferred-term substitutions. Returns issues for violations plus suggested from->to replacements.

POST /calculate $0.07

Independently re-compute order/invoice math from line items and optional fees, returning gross, fee and net so an agent's own totals can be verified against a trusted calculation.

POST /knowledge $0.07

Check a generated answer against a set of approved facts and forbidden claims to catch ungrounded or off-limits statements. Flags forbidden claims found in the answer and warns when the answer matches none of the approved facts.

POST /policy $0.07

Check an intended action — vendor, amount and category — against a spend policy (max amount, allowed/blocked vendors, allowed/blocked categories) before money moves. Returns allowed=false with reasons when any rule is violated.

POST /schema $0.07

Validate that a structured output contains the required fields and that each field has the expected JSON type. Returns valid=false plus a list of errors when a field is missing or mistyped.

View External Documentation →