AI
Consumption Control
Cap AI-agent usage and cost — per agent, per user, per organisation. Prevents invoice surprises and steers where AI budget actually goes.
Updated May 18, 2026
Configuration · AI · 8.4
Consumption Control caps AI usage in Gfacility — per agent, per user, per organisation or per month. AI costs are variable; without limits, a runaway agent or an enthusiastic user can eat your AI budget in a week. Here you set the limits the system enforces itself.
Why this matters to the business
"AI invoice 3× expectation"
Monthly budget per tenant → hard cap prevents the runaway scenario.
"One user burns it all"
Per-user limit → fair share, no one-user-monopolises scenario.
"Pilot and production same quota"
Different limits per agent — experimental agent small, production agent generous.
"No visibility where money goes"
Consumption dashboard per agent/user → decisions on numbers, not on gut feel.
Levels of limits
Tenant
Monthly budget for the whole organisation. Hard cap; on overrun all agents pause.
Agent
Limit per AI Agent — every agent gets its own budget inside the tenant total.
User
Maximum requests per user per day/week. Prevents one person consuming disproportionately.
Organisation / department
Multi-tenant or chargeback? Budget per department — breakdown for finance.
What you set per limit
| Field | Example |
|---|---|
| Unit | Requests · Input tokens · Output tokens · Euros. |
| Period | Day · Week · Month · Calendar month. Auto-reset. |
| Hard / soft cap | Hard = stop immediately; soft = warn but let through. |
| Alert thresholds | At 50%, 80%, 100% — notification to admin. |
| Behaviour on overrun | Pause agent · Request approval · Route to a human. |
Which decisions will you make?
Tenant monthly budget
Start conservative, scale on real usage. Measure 3 months first, then fix a budget.
Hard vs soft cap
Production agents hard (known behaviour); experimental soft (let it break, learn from it).
Chargeback per department?
Multi-org and chargeback active? Budget per department with monthly report — feeds the finance export.
Alert routes
Who gets the 80% warning? IT-platform admin + AI owner + finance.