The core of the Xybern Control Plane. Every AI output, from any model, any agent, any format, is decomposed into claims, mapped to evidence, and scored deterministically.
Three pillars of the verification engine.
Every AI output is broken into individual, verifiable claims. Each claim is independently scored against source evidence, no black-box assessments.
An immutable records layer that tracks every data access and reasoning step. Every output is anchored to original evidence, not model training data.
Zero LLM variance. Our proprietary scoring model produces identical results for identical inputs. Every score is reproducible, auditable, and permanently recorded.
Send any AI output to POST /api/v1/verify and get back a deterministic trust score, claim-level breakdown, and a permanent vault record.
One API call covers verification, governance, and vault recording. No orchestration needed.
Every response includes trust score, individual claims, evidence mapping, and governance outcome.
Scoped API keys with workspace isolation, rate limiting, and full access logging.
// POST /api/v1/verify { "content": "The contract contains a 90-day termination clause...", "source": { "type": "llm", "model": "gpt-4", "provider": "openai" }, "original_prompt": "Analyze this contract" } // Response { "ok": true, "verification_id": "ver_a8f3c9e01b24", "trust_score": 78, "claims": [ { "text": "90-day termination clause", "score": 92 }, { "text": "Standard commercial terms", "score": 64 } ], "governance": { "outcome": "pass" }, "vault_hash": "sha256:e3b0c44..." }
The Verification Engine is built for enterprise AI infrastructure where governance is mandatory. Every verification is cryptographically anchored and fully auditable.
Rigorous external auditing of all security controls.
Enterprise identity and granular access logs.
Strict logical separation of customer environments.
Every data touchpoint recorded in the Vault.
