AI Configuration
AI configuration is the admin side of Fesk's AI, where you set guardrails and review workflow proposals and actions. The end-user features themselves are covered in AI features.
All of this lives under Settings → AI and requires an admin role.
How AI settings are scoped
AI settings apply per project. Each project gets its own AI configuration automatically, there's nothing to create or install. You control a project's behavior from Settings → AI: its guardrails and its spending cap. Switch projects to configure another one.
Where AI acts
There is no project-wide autonomy dial. AI acts only where a workflow step is configured to, and each step sets how far it can go on its own:
- Suggest: AI only proposes; a human approves every action.
- Semi: AI acts immediately, and the action lands in the audit queue for post-hoc review.
- Auto: AI acts without mandatory review.
You configure these per step when you build a workflow (see AI workflows). Everything a step does autonomously is still bounded by the guardrails, spending cap, and kill switch below. A threshold preview under Settings → AI shows how a given confidence threshold would have applied to recent tickets, so you can tune guardrails before trusting them.
Guardrails
Guardrails constrain autonomous behavior, confidence thresholds, which actions are permitted, and where AI must always defer to a person. Set these conservatively at first and loosen them as you build trust.
Audit queue and kill switch
Everything AI does autonomously is recorded in the audit queue so you can check its work:
- Filter the queue by Pending review to see proposals or executed Semi-mode actions awaiting review. Approving a proposal executes it after a fresh safety check; acknowledging an executed action confirms it. Eligible executed actions can be reverted. The action-type filter covers reply, close, assignment, escalation, and status changes.
- The kill switch immediately halts all autonomous AI action workspace-wide. Hit it if something looks wrong; clear it when you're ready to resume.
Cost control
AI usage costs money, so set a monthly AI cost limit (under billing). When usage hits the cap, automated AI work pauses until the next period or until you raise the limit, it never silently overspends.
Related
- What these features do for agents and customers: AI features.
- Usage metering and the cost limit: billing.