Last updated: July 23, 2026
From requests to accountable usage
Kimss ties every model and agent request to a tenant, workspace, agent, and billing key—then records consumption as governed requests your finance and engineering teams can share.
Governed-request governance begins at attribution. Without knowing which workspace and agent drove consumption, allowances and caps are guesswork. The universal gateway on /v1/agents/run and /v1/models/completions attaches identity and metering before the vaulted provider executes.
Attribute
Associate requests with tenant, workspace, agent, and billing key for every governed call.
Allocate
Set monthly workspace allowances and optional group-level governed-request budgets inside the workspace.
Respond
Use soft alerts and explicit exhaustion policies before consumption becomes an incident.
A governance model teams can operate
governed-request allowances answer how much is available; usage aggregates answer what was consumed; the append-only billing ledger preserves allocations, top-ups, overage, and adjustments for reconciliation.
Group budgets add a soft allocation layer for departments without changing the tenant-wide enforcement boundary when the workspace allowance is exhausted. Operators configure thresholds so product teams receive warnings before hard blocks.
Governed requests normalize usage for product governance. They do not replace the underlying Azure invoice or claim identical economics across every model tier.
Operational playbooks
Run monthly reviews comparing allowance sizes to usage trends, adjust group budgets after launches, and investigate spikes using agent-level attribution before expanding provider capacity.
Platform teams publish internal runbooks: who approves allowance increases, how top-ups appear in the ledger, and which roles may view usage versus change caps. Pair Kimss surfaces with Azure Cost Management for infrastructure truth.
Developers should prefer /v1 routes so attribution matches current product behavior; see /assistants-to-v1-migration if legacy assistant calls remain.
Connect budgets to execution
Governed requests are enforced at the gateway before traffic reaches your vaulted provider—so an exhausted allowance actually stops governed calls instead of merely reporting after the fact.
See /docs/usage_credits for metering, spend cap concepts at /ai-spend-caps-and-credits, and enterprise onboarding at /enterprise when contractual allowances differ from self-serve plans at /pricing.