Pre-call inspection
Inspect text inputs before the selected provider receives them, including text parts inside multimodal requests.
USE CASE / DATA PROTECTION
CONTROL OUTCOMES
Inspect text inputs before the selected provider receives them, including text parts inside multimodal requests.
Evaluate model output before it is returned to the calling application.
Configure allow, flag, redact, or block behavior by detector, direction, and governance scope.
Persist guardrail decisions with the virtual key, provider, model, usage, and final request outcome.
REFERENCE ARCHITECTURE
Resolve the virtual key and its organization, team, project, and applicable policy scopes.
Run configured PII, secret, and denylist detectors before routing upstream.
Call an approved deployment, then apply response-direction guardrails before returning output.
Write the call and guardrail events to the durable audit record and optional SIEM projection.
TRUST ARCHITECTURE
Map virtual-key access, encrypted provider credentials, guardrail decisions, and audit evidence to the controls your enterprise already operates.
Architecture supports evidence collection for access, change, and monitoring controls.
Map credential, access, policy, and audit practices to ISMS control objectives.
Operational visibility across governed model access, guardrails, usage, and provider outcomes.
Control-alignment statements describe product architecture and are not claims of EnvisionAI certification.
TECHNICAL EVALUATION
Map the architecture to your providers, application clients, governance scopes, data boundaries, and operating requirements.