DCL Trust Oracle

Low-latency pre-action screening for routine agent responses and tool-call decisions.

DCL Trust Oracle. Call only when the text must include the substring AI and must not include "I cannot be held responsible" or "no guarantees". POST JSON {"response","agent_id"}. COMMIT means those checks passed and the action may proceed. NO_COMMIT means do not execute the checked action. Not a tool-call guardrail, prompt-injection detector, or jailbreak screen; use /evaluate/fast for the phrase check.

Detect jailbreaks, prompt injection, and instruction conflicts before the agent follows them.

Scan text for personal data: email, phone, national id, bank card, IBAN, crypto address, IP address, passport. COMMIT when nothing matches. NO_COMMIT on any finding. POST JSON {"response","agent_id"}.

Scan text for exposed credentials, API keys, private keys, database URLs, and webhook secrets. COMMIT when nothing matches. NO_COMMIT on any finding. POST JSON {"response","agent_id"}.

Read one audit-chain record by tx_hash, plus drift context, chain position and tamper location. Use the tx_hash returned by any earlier DCL evaluate or scan call. GET path parameter, no JSON body.

Evaluate a list of texts in one paid call. Each item may name a built-in policy. The call is rejected when the list is longer than max_items (default 20). POST JSON {"items","agent_id","max_items"}.

Check agent outputs for quality issues, inconsistency, and behavioral drift.

Higher-confidence evaluation for high-impact, sensitive, or difficult-to-reverse agent actions.

Read one audit-chain record by tx_hash. GET path parameter, no JSON body. Returns verdict, confidence, agent_id, and reason. Does not return raw content.

Jailbreak / instruction-adherence detection.

Content quality and drift check.

Baseline safety policy check.

Strict pre-action audit with a higher confidence bar.