AI Features

Fesk's AI assistant, Fesky, drafts replies for agents and runs explicitly configured workflow steps within rules you define.

Suggest Reply

On any ticket, the Suggest Reply button drafts a response based on the ticket's content, its conversation history, and relevant published knowledge-base articles. An agent reads it, edits if needed, and sends, or scraps it. Nothing goes to the customer unless an agent clicks send.

Suggest Reply doesn't change the ticket's state on its own. It's a writing aid, not an actor.

Autonomous actions

There is no workspace-wide autonomous mode. AI takes real actions — replying, closing, routing — only where a workflow step is configured to, and each step's mode sets how far it goes on its own:

• Suggest (the default for new resolve steps): the AI proposes an allowed action, you review and approve it. Nothing happens without a human click.
• Semi: the step executes on its own, and the action lands in the audit queue for post-hoc review.
• Auto: the step executes without mandatory review. Guardrail hits and low-confidence cases still queue.

New projects include an Auto triage step that applies high-confidence category and priority predictions; resolve steps must be added and authorized explicitly. Each resolve step has its own action allowlist. Steps are configured in the workflow builder (see AI workflows). Proposed and executed resolve actions are recorded in the audit queue.

Guardrails

In Settings → AI, admins set what the AI can and can't act on:

• Block configured sentiments from automatic closing
• Require configured confidence, knowledge-base match, and intent-match thresholds
• Cap automatic closes per hour
• Restrict autonomous actions by language and exclude specific ticket types

Audit queue

Every proposed or executed resolve action is logged in Settings → AI → Audit Queue. Team leads and admins can approve proposals, acknowledge executed actions, and revert eligible mutations. Revert restores supported ticket fields and can remove an AI reply when no later customer reply makes that unsafe; it cannot retract a notification that was already delivered.

The kill switch

Use Settings → AI → Kill Switch to stop autonomous workflow processing. Pending workflow and autonomous tasks are cancelled where possible; callbacks dispatched before the kill remain invalid even after processing is re-enabled. The manual Suggest Reply writing aid is controlled by AI availability and cost limits, not by the autonomy kill switch.

What the AI sees

AI requests are scoped to the current workspace and can include the ticket, its conversation, and relevant published knowledge-base content. Fesk's published data policy states that this content is not used for model training. To stop autonomous actions, use the kill switch; to cap or disable paid AI usage, use the AI cost controls in billing settings.