FAQ

Are documents and queries sent to third-party model providers?

Only if the deployment is configured to use an external model provider for that request path. Administrators can choose model providers and should document what prompts, retrieved context, files, and metadata may leave the deployment.

Is data used for model training?

AIvis does not train models from customer data as part of normal product operation. External provider behavior depends on the selected provider and contract. Review provider terms before sending sensitive data to any external model service.

Where is data stored?

Data is stored in the deployment’s configured Postgres, OpenSearch, object storage, Redis, and logging systems. See Data Storage.

Can AIvis run in a private or offline environment?

AIvis can be deployed with private infrastructure paths. Fully offline operation depends on whether the deployment also provides local model services and avoids external connectors, web search, or third-party processors.

How are documents protected?

Documents are protected through connector scope, document visibility settings, groups, document sets, and retrieval-time authorization. Source-system permissions should be tested for each connector path before production rollout.

How are Agents controlled?

Agents are controlled through their configuration, allowed knowledge, allowed models, tool allowlists, and AIvis service-side authorization. Prompt instructions do not replace authorization checks.

Can an Agent leak confidential information through telemetry or logs?

Agent telemetry should store usage facts such as request IDs, model identifiers, token counts, status, and cost snapshots. It should not store full prompts, model responses, reasoning traces, raw tool parameters, or secret values. See Agent Security Strategy.

Can a prompt override document permissions or tool policy?

No. Prompt text and Agent instructions are behavior guidance, not authorization. Document retrieval, tool execution, model availability, and administrative actions must be enforced by AIvis services.

What should be logged?

Log request IDs, authorization denials, connector failures, indexing jobs, model errors, runtime failures, and administrative changes. Avoid logging raw credentials, full secret payloads, or unrestricted sensitive content.

What must deployment owners manage?

Deployment owners manage TLS, firewall rules, network segmentation, secret rotation, identity-provider configuration, backups, restore testing, host hardening, monitoring, incident response, and regulatory controls for their environment.