Skip to main content
AI Assistant processes Zendesk data so it can retrieve the right knowledge, call approved systems, generate a response, and provide operational evidence afterward. The exact fields involved depend on your Zendesk configuration, enabled features, connected tools, and the operation an agent or webhook invokes.

​
Typical Assistant request flow

1

Zendesk sends ticket context to RAUM

The Zendesk app or configured webhook sends the context needed for the requested Assistant operation.
2

RAUM loads organization configuration

RAUM resolves the organization’s models, applicable procedures and guidelines, Zendesk metadata, knowledge sources, and tool configuration.
3

RAUM retrieves evidence

Semantic knowledge retrieval and configured SOURCE tools can provide information required to answer the ticket.
4

Personal information is masked for AI processing

Supported personal/sensitive values are replaced with placeholders in AI requests and relevant AI logs before model processing.
5

RAUM calls your AI provider

The required prompt/context is sent using the AI-provider credentials configured for your organization.
6

The result returns through RAUM

RAUM applies the response workflow, restores supported masked values where necessary, evaluates confidence/action state, and returns or delivers the result to Zendesk.

​
What RAUM may store

The platform stores the organization resources required to operate the service, which can include configuration and support-processing records such as:
  • user and organization settings;
  • encrypted provider/Zendesk/integration credentials;
  • procedures and AI Guidelines;
  • synchronized/indexed knowledge and embeddings;
  • model configuration;
  • Zendesk metadata used by configured features;
  • Assistant activity/system/audit records; and
  • generated operational data required by enabled features.
Do not rely on old documentation that claimed raw ticket data is categorically never stored by RAUM. The current platform maintains operational records needed for features such as Activity, quality analysis, retrieval, and debugging/audit evidence.

​
AI-provider data

Because RAUM is BYOK, the provider receives model requests under your provider account, subject to the provider terms and data controls attached to that account. Your provider choice therefore matters for:
  • model availability;
  • region;
  • provider-side retention/training policy;
  • enterprise data controls; and
  • contractual terms.
Review those settings directly with the provider you connect.

​
Credentials

Sensitive credentials are encrypted by the application with AES-256-GCM before database storage. Encryption key material is kept separately through Azure Key Vault. RAUM does not return stored secrets back to normal browser views in plaintext after they have been saved. Where possible, use the Connections Authentication tab to store a reusable encrypted credential and reference it from integrations rather than creating several copies.

​
Logs and Activity

RAUM keeps operational evidence needed to understand Assistant behavior. Ticket Activity can include the result and metadata needed to diagnose which procedures/guidelines/tools participated in the run. System Activity and Audit record administrative/operational events. Personal-data masking is applied to AI/logging paths as part of the security design, but customers should still minimize unnecessary sensitive information in authored procedures, guidelines, and external tool outputs.

​
Attachments and external sources

When attachment processing, Gallery, documents, websites, or external API/MCP tools are enabled, RAUM processes the relevant content needed for those features. Treat every connected source as part of your support-data boundary. A third-party connection can expose data to RAUM that Zendesk itself never contained.

​
Retention and deletion

Retention behavior can differ by resource and product feature. Avoid using an old fixed-duration claim as a universal RAUM retention rule. Use the current RAUM privacy/legal documentation and your applicable service agreement for authoritative retention and deletion commitments. For AI Chatbot conversations, separate configurable transcript retention/deletion controls apply and are documented in the Chatbot section.

​
Data-minimization recommendations

  • Do not place credentials or secret tokens in procedures or AI Guidelines.
  • Avoid sending sensitive data to a tool unless the workflow needs it.
  • Use SOURCE tools to retrieve only the fields required for the answer.
  • Restrict ACTION inputs and approvals.
  • Remove stale knowledge and inactive integrations.
  • Review provider-side data settings for the BYOK account.
  • Test new workflows in Simulation before allowing side effects.

Security overview

Review encryption, provider flow, action safety, and auditability.

Activity and audit

Inspect the evidence RAUM records for Assistant work.