FinOps and cost governance
ZeaLLM treats AI spend governance as a platform capability spanning the gateway and portal. Every priced request is attributed at ingestion time, checked against applicable budgets, and retained as request-level evidence for reporting and allocation.
For a detailed entity map, see FinOps relationship and data flow.
Attribution
Use managed FinOps dimensions for departments, business units, projects, clients, environments, and custom business metadata. Values can be hierarchical and assigned to organizations, teams, applications, keys, users, customers, cost centers, and agents with effective dates. Managed cost centers are separate first-class allocation records, not tags or custom dimension values.
Required-metadata policies can reject requests that would otherwise create unallocated spend. Spend records retain a metadata snapshot so later hierarchy or ownership changes do not rewrite historical attribution.
Cost events extend metering beyond chat and embeddings. Tools, external APIs, compute, storage, network, and adjustments use the same attribution fields as model calls.
Agent and tool costs
Agents, tools, runs, and tool calls are first-class records. An agent run combines model, tool, and infrastructure costs and carries organization, team, application, cost-center, environment, user, and managed-dimension attribution.
Agent and tool ingestion APIs should use scoped credentials and stable external identifiers. Do not place prompts, credentials, or customer payloads in cost event metadata.
Reports and drill-through
Spend rollup generation defaults to hourly, daily, and monthly aggregates. Quarterly is a supported rollup granularity, but it must be requested explicitly rather than being generated by default.
FinOps Analytics reports can group by their built-in dimensions: model, provider, user, team, application, organization, customer, cost center, environment, deployment, and agent. Governed custom dimensions are retained in each record's snapshot and managed in FinOps Attribution, but are not dynamically added to the current FinOps Analytics Attribute by options.
Every aggregate preserves a path to filtered request logs and cost events. Scheduled reports record each delivery attempt and its reporting window.
Chargeback and showback
A rate card converts provider cost into billed cost using effective-dated prices, discounts, and markups. Allocation policies distribute shared cost directly, equally, by percentage, or in proportion to measured usage.
Closing a period creates snapshotted statements and line items. Publish statements for showback, or use billed totals for chargeback. Provider invoice details can be entered and reconciled against metered provider cost; investigate material variances before publishing a period.
Approvals and overrides
Governance policies can require approval for budget increases, high-cost models, high-cost agents, and emergency overrides. Policies identify approvers, thresholds, expiry, service-level targets, and escalation recipients.
Emergency overrides must always include a reason and expiry. Their grant, revocation, and use are auditable.
Optimization
Optimization analysis detects unpriced traffic, missing attribution, spend anomalies, idle keys, unused deployments, duplicate agents, and cheaper-model opportunities. Recommendations include evidence, estimated savings, and confidence.
A lower list price alone does not prove equivalent quality. Model-switching recommendations remain proposals until supported by workload-specific evaluation evidence and approved routing constraints.