AI usage without visibility: unknown consumption and scattered keys with no audit

A store owner enables an AI assistant for the team, then faces surprising consumption with no view of which prompts failed or which tools executed, and API keys scattered with no clear owner. The problem is governance, not rejection of AI.

Why this happens

Because chat consumes tokens with every message and tool call, keys get created per experiment with no central record, and executions happen without a reviewable trail.

What Vareons does about it

Vareons provides the /dashboard/ai-usage page with token summaries, prompt logs, CSV export, and a 20-execution audit table, plus the /dashboard/ai-usage/api-keys page with tenant-scoped keys encrypted via AES-256-GCM envelope with HMAC and masked previews only (per the facts registry). Every change passes an explicit confirmation gate.

ai.copilot.usage.dashboard.supported — AI usage dashboard: Available
ai.copilot.usage.tool-audit.supported — AI tool audit log: Available
ai.copilot.keys.management.supported — AI provider key management: Available
ai.copilot.mutations.confirmation_required — AI Copilot mutation confirmation gate: Available

Limitations

The audit table shows only the last 20 executions, export covers 30 days, distribution covers seven days. The page reports usage; it does not itself enforce an automatic cap.

Last verified: 2026-09-19