Skip to main content

Budget & Cost Management

Rivaro tracks the cost of every AI action and enforces budget limits as part of the governance chain. Budget enforcement is not a reporting layer bolted on after the fact — it is part of the enforcement chain and can WARN, THROTTLE, or BLOCK actions before they execute.

Where to find it in the app​

Dashboard → Administration → Budgets.

The Budgets page has two tabs:

  • Budget Policies — create, edit, and toggle budget policies at org, department, and agent scope. This is where you set the limits and choose warn/throttle/block thresholds.
  • Model Pricing — Rivaro's per-model pricing table (what each provider charges per 1K tokens). Used for cost estimation. Read-mostly, with override capability for self-hosted or contract-discounted models.

The page header has an Export costs dropdown for CSV or FOCUS-format cost exports.

Cost rollups are also surfaced on:

  • Command Center — Cost Intelligence widget summarizing org-wide spend
  • Executive View → Reports — Cost Overview narrative report
  • Agent Registry → agent → Activity — per-agent cost timeline
  • Actions & Evidence → Actions & Outcomes — per-action cost on each row

Budget Policies​

Budget policies define spend limits at three hierarchical scopes:

ScopeWhat it controlsExample
OrganizationTotal AI spend across the entire org"$50,000/month org-wide"
DepartmentSpend for a business unit or team"$10,000/month for the Claims team"
Agent (AppContext)Spend for a single agent or integration"$500/month for the Customer Support agent"

Create policies from Administration → Budgets → Budget Policies → Add policy. Pick the scope, name the target, set the period and limit, then set the warn / throttle / block thresholds.

Hierarchy​

Budgets are hierarchical — a child scope cannot exceed its parent:

  • An agent budget cannot exceed its department budget
  • A department budget cannot exceed the org budget
  • If no agent-level budget exists, the department or org budget applies

When multiple budgets apply to a single action, the strictest action wins. If the org budget says WARN but the department budget says BLOCK, the action is blocked.

Budget actions​

When spend approaches or exceeds a budget threshold:

ActionWhen
WARNSpend exceeds a warning threshold (e.g. 80% of budget). Action proceeds, alert generated.
THROTTLESpend exceeds a throttle threshold. Action rate is reduced.
BLOCKSpend exceeds the hard limit. Action is rejected before it executes.

Each policy lets you set the percentage threshold for each action independently. A common pattern: WARN at 80%, THROTTLE at 95%, BLOCK at 100%.

How budget enforcement works​

Budget is one of the checks Rivaro runs before a tool call executes. On every tool call:

  1. Rivaro looks up all applicable budget policies (org, department, agent)
  2. Current spend is evaluated against each policy's thresholds
  3. The strictest action across all applicable scopes is returned
  4. If the result is BLOCK, the tool call is rejected before execution

Budget evaluation runs on a hot path — spend totals are pre-computed and cached so that budget checks add minimal latency to the enforcement chain.

Cost tracking​

Every request through the Rivaro Gateway and every tool call through the sidecar carries a cost estimate:

  • Token costs — estimated from model, provider, and token count using the rates on the Model Pricing tab
  • Per-request cost — attached to every action record and telemetry metric
  • Cumulative spend — rolled up by agent, department, organization, session, and arbitrary tags

Cost data is captured on every request regardless of whether budget policies are configured. You get spend visibility even before you set limits.

Cost Intelligence​

The cost intelligence layer provides reporting across the agent estate. The Command Center surfaces a summary widget; the Executive View → Reports tab has the full Cost Overview narrative report. The same data is exportable via the Export costs action on the Budgets page.

ReportWhat it shows
Organization overviewTotal spend, spend by provider, spend by model, trends
Agent cost attributionPer-agent spend, cost per action, cost trends over time
Avoided spendCost of actions that were blocked by policy — spend that would have occurred without governance
Cost by decisionSpend broken down by enforcement decision (ALLOW, BLOCK, DEFER, etc.)
Session costsPer-session cost rollup for debugging and optimization
Tag-based rollupsCost grouped by arbitrary tags (project, use case, environment, etc.)

Model pricing overrides​

If you have a negotiated contract rate with a provider, or you self-host an OpenAI-compatible model with your own cost basis, open Administration → Budgets → Model Pricing, find the model row, and click Override to set your own price-per-1K-tokens. Cost estimates and budget enforcement immediately use your override for that model.

Next steps​