What to register
Register the agent's stable technical identity, intended purpose, responsible organizational function, provider and model version, environment, autonomy, lifecycle, risk tier, data classes, declared tools, access level and review expiry. Do not enter names, emails, credentials, prompts, conversations or tool payloads.
Eight evidence dimensions
The model covers identity and accountability; authorization and delegation; tools and supply chain; evaluation, reliability and safety; human oversight; audit and observability; privacy and data governance; and lifecycle and incident readiness. Each applicable control needs a status, measurement window, expiry, evidence class and artifact digest.
Interpret the result
Trust score, evidence coverage, freshness, confidence and independence are separate. A critical failure restricts the disposition. Missing critical dimensions or low coverage produces insufficient_evidence rather than an optimistic number. Reassess after a model, tool, permission, policy or deployment-context change.
KPIs that matter
Use deployment-representative task success and safe completion, policy violations, unauthorized tool calls, attack success, tool precision and retries, loop termination, human approval and override, grounding error, latency, cost per successful task, incidents and recovery time. Every KPI needs a numerator, denominator, sample window, context and minimum sample. Token volume and raw invocation count are diagnostic only.
Privacy, support and current boundary
Read the product privacy notice, service-provider register and security policy. The current release is a controlled pilot without a public availability SLA or certification claim. Contact support@qentra.tech for support or deletion assistance.