Last updated: August 23, 2026
Why governed requests beat raw tokens for governance
governed requests normalize usage across models into one allocatable unit—so finance sets monthly allowances and engineering still chooses the best model per task.
Token pricing varies by model and changes over time. Product governance needs a stable vocabulary. Governed requests translate consumption into workspace allowances, group budgets, thresholds, and ledger entries finance can reconcile alongside Azure invoices.
Governed requests govern product behavior; they do not claim to replace Azure billing or make every model economically identical.
Workspace allowances and group budgets
Each workspace receives a monthly governed-request allowance; optional group budgets subdivide that allowance for teams without breaking tenant-wide enforcement when the workspace exhausts capacity.
Soft alerts warn owners as consumption approaches limits. Exhaustion policies determine whether requests hard-stop or require manual top-up—configuration your operators choose per environment.
Append-only billing ledgers record allocations, usage, top-ups, overage, and adjustments for audits and chargeback.
Attribution down to agents and keys
Usage aggregates tie requests to tenant, workspace, agent, and billing key—so overrun investigations identify the workload, not just the invoice line.
Attribution flows through the universal gateway on /v1 routes. Legacy assistant paths remain metered where mounted, but new integrations should prefer /v1 for consistent behavior.
See also /ai-credit-governance for a focused narrative on team operations.
Per-endpoint token caps (Provider Vault)
Govern workspace allowances with governed requests—and cap runaway provider token spend on individual vaulted models without slowing down engineering.
On Production, Scale, and Enterprise, workspace admins can set an optional monthly token cap per Connected Infrastructure endpoint. Choose email alerts at 50/80/90/100% or hard-block with HTTP 429 when the cap is exhausted. This meters your provider tokens—not Kimss governed requests—so finance can stop a single fine-tuned model from blowing the Azure invoice while the rest of the workspace keeps running.
See Connected Infrastructure and Usage & governed requests for the two-meter model.
Connecting caps to platform decisions
Spend caps let platform teams approve new agent workloads by adjusting allowances instead of provisioning unbounded Foundry capacity—aligning AI growth with budget cycles.
Review plans at /pricing and enterprise options at /enterprise when monthly allowances need contractual adjustment or dedicated capacity.