AI agents and autonomy

The roster

Arthiva ships a roster of specialised agents that watch your operation continuously: duplicate-return detection, fraud and abuse signals, SLA breach prediction, stale-RMA nudges, warranty verification, return forecasting, photo inspection, low-stock alerts, vendor news, and more. Org admins manage them under AI → Agents — each agent is enabled per workspace, individually.

The autonomy dial

Each agent's Behavior setting answers one question: how much may it do without a human?

• Insight only — the agent raises findings; people act.
• Propose — the agent drafts the action (e.g. cancel a duplicate RMA) and queues it for your Approve, optionally notifying the right people.
• Auto-approve safe actions — routine, low-risk cases run on their own; everything else still waits for a human.

Start on Insight only and promote an agent when its judgement has earned it. Every proposed action shows its reasoning before you approve, and the nightly analysis (02:00 UTC) means findings are waiting for you each morning.

Where the work shows up

AI → Activity is the feed of everything agents have found, each entry with a plain-language explanation of why — no black-box verdicts. AI → Overview is the scoreboard: what each agent caught, how often its calls were confirmed, and the measured value — hours saved and losses avoided — so you can judge every agent by results, not promises.