Agent Mode

Agent Mode is AI Assistant’s webhook-driven autonomous operating mode. Instead of waiting for an agent to click Generate, RAUM processes qualifying Zendesk ticket events in the background and applies the delivery policy configured for the organization.

​

Enable Agent Mode

Open: AI Assistant → Zendesk → Automation → AI Mode Choose Agent Mode and configure how RAUM should deliver results.

​

Internal-note-only mode

Enable Add response as internal note when you want the Agent pipeline to run automatically but still require an agent to send the customer-facing response. This is a useful intermediate rollout stage between Copilot and fully public autonomous replies.

​

Public replies by confidence

When internal-note-only mode is off, choose which confidence bands are allowed to receive a public reply:

• High
• Medium
• Low

Confidence bands you do not select are handled as internal notes rather than public customer replies. Example: if you allow only High, Medium and Low results still run through Agent Mode but are surfaced internally for human handling.

​

Ticket completion

Agent Mode also includes a fallback setting for tickets where the AI did not explicitly decide a Zendesk status. When Solve tickets the AI did not decide a status for is enabled, RAUM can solve a ticket after a qualifying public reply if no procedure or AI decision already chose another status. Explicit status decisions from the pipeline or procedure directives take precedence.

​

Required field completion

Agent Mode can require the expected Zendesk fields to be completed before the autonomous workflow is considered complete. Use this when downstream reporting, routing, or business processes depend on specific field values rather than only the customer reply.

​

Webhook processing

Agent Mode depends on Zendesk events reaching RAUM. Onboarding can automatically provision the normal webhook and triggers. Administrators can later review automation under: AI Assistant → Zendesk → Automation → Webhooks The onboarding trigger set includes:

• Ticket created
• Ticket solved
• Ticket rated

You can also use Zendesk trigger conditions to narrow which tickets send the events RAUM uses.

​

Skip rules

The Automation configuration can exclude tickets from AI handling based on configured criteria such as:

• phrases;
• regular expressions;
• Zendesk channels;
• tags;
• organizations; and
• groups.

Use explicit skip rules for cases where RAUM should never participate rather than relying on a procedure to tell the AI to ignore a ticket after processing has already begun.

​

SOURCE and ACTION tools

Autonomous processing follows the same tool safety model as Copilot:

• SOURCE tools retrieve information.
• ACTION tools can change state.

ACTION tools remain subject to their configured approval/authorization behavior. Do not convert state-changing calls to SOURCE to make Agent Mode appear more autonomous.

​

Simulation before production

Open: Quality & Testing → Simulation Simulation runs the full Assistant pipeline against a Zendesk ticket.

• SOURCE tools execute normally so the test has real retrieval context.
• ACTION tools are intercepted and returned as previews without performing their side effects.

Use Simulation after changes to:

• procedures;
• models;
• connections;
• ACTION approval behavior; or
• major AI Guidelines.

​

Monitor Agent Mode

After enabling automation, review:

• Dashboard → Autonomy & deflection
• Dashboard → Answer quality
• Dashboard → Latency & reliability
• Activity → Ticket Activity
• Quality & Testing → CSAT & Outcomes
• Quality & Testing → Knowledge Gaps

These views provide more useful evidence than relying on the fact that tickets happened to close successfully.

​

Recommended rollout