Get started

Concepts

Key terminology used across the Xobriq platform.

Risk score

A floating point between 0 and 1 returned by Guard. Represents the estimated probability that an event is fraudulent.

Signal

A single feature contributing to the risk score — velocity, geolocation, device fingerprint, behavioural biometrics, etc. Xobriq Guard uses 120+ signals.

Decision

A recommended action: allow, review, block, or step_up. Determined by the risk score and your configured thresholds.

Agent

An autonomous LLM-powered worker that executes a multi-step workflow. Xobriq ships pre-built agents for Fraud Investigation, KYC, Compliance, and SecOps.

Sovereign partition

A dedicated slice of H200 GPUs on Xobriq Cloud with hardware-level isolation, guaranteed data residency, and air-gapped deployment options for government workloads.

Public endpoint vs. dedicated endpoint

Public endpoints are pre-deployed models shared across tenants — perfect for prototyping. Dedicated endpoints give you a private model instance with reserved throughput.

Environment naming

Sandbox keys start with sk_test_ and hit the sandbox environment. Production keys start with sk_live_ and hit the production environment.