One-call KYB for AI agents: vet any company

Last updated: May 21, 2026

Know Your Business (KYB) means verifying a company is real, legitimate, and safe to transact with. For an AI agent, the cleanest pattern is three composite tools over public records: resolve_entity to pin the identity, company_trust_check for a standing trust read, and counterparty_risk_score for a weighted 0-100 verdict. Each is one call, and each fans out across SEC, sanctions, courts, EPA, and federal spending for you.

Step 1: resolve the identity

Company names are ambiguous, so start by resolving the name to canonical identifiers before anything else. resolve_entity returns the SEC CIK and ticker, GLEIF LEI (plus immediate parent), IRS EIN, USAspending recipient, and EPA registry id, each with a per-source confidence.

{
  "method": "tools/call",
  "params": {
    "name": "resolve_entity",
    "arguments": { "name": "Acme Robotics Inc" }
  }
}

Now downstream tools work off a real join key instead of fuzzy-matching the name again.

Step 2: the standing trust read

company_trust_check answers "can I trust this company?" in one call. It runs an OFAC, EU, UN, and BIS sanctions screen (a hit is decisive), checks SEC EDGAR registration and the latest filing for legitimacy and recency, looks for USAspending federal awards, and pulls EPA compliance flags, then rolls it into a verdict band.

{
  "method": "tools/call",
  "params": {
    "name": "company_trust_check",
    "arguments": { "company": "Acme Robotics Inc" }
  }
}

Step 3: the weighted risk score

When you need a single number to gate a workflow, counterparty_risk_score combines sanctions, SEC EDGAR, federal court litigation, EPA enforcement, and federal spending into one weighted, explainable 0-100 metric with an evidence chain, so the agent can show its work.

{
  "method": "tools/call",
  "params": {
    "name": "counterparty_risk_score",
    "arguments": { "company_name": "Acme Robotics Inc" }
  }
}

What this is and is not

These tools synthesize public records into a signal read. They are not a credit report, a background check, or an endorsement, and every leg is best-effort: an unavailable source is reported, never silently dropped. Use the verdict to triage and to decide where a human should look harder, not as the sole basis for a legal or lending decision. The packaged version of this workflow, with the pricing and coverage detail, lives at /compliance-pack.

FAQ

Why resolve the entity first?

A name like "Acme" can match dozens of filers. Resolving to a CIK, LEI, or EIN first means the trust check and risk score run against the right company instead of guessing, which cuts false matches on every downstream leg.

Does a sanctions hit mean I must stop?

Treat it as a decisive flag to investigate, not an automatic verdict. Sanctions screening is fuzzy name matching, so confirm the candidate against identifiers before you act, and consult compliance or counsel on a true match.

Can the agent do this in one turn?

Yes. All three are single tool calls, and a well-prompted agent can chain resolve, trust check, and risk score in sequence, then summarize the evidence. See /decisions for the full set of one-call decision tools.

Last updated: May 21, 2026 · LiveDataLink home