Spend governance
See what every agent and model is costing you over time, set a workspace budget, and act on cost from the same gateway that already governs risk.
AxioRank already sees every governed call. Spend governance turns that stream into
a live cost view: report the model and token counts (or a costUsd) on your
gateway calls and AxioRank attributes spend per agent and per model, charts it over
time, and lets you cap it. It is cost control sitting in the same control plane as
risk, not a separate billing tool.
Per-agent budgets vs. workspace spend
Looking for a hard cap on a single runaway agent's calls or cost? That is a budget policy, enforced inline per agent. This page is the workspace-wide view: total spend, the per-model and per-agent breakdown, and the one monthly budget for the whole workspace.
The spend dashboard
The /spend page shows, for your reporting window:
- Headline totals: month-to-date spend, spend over the last N days, governed call count, and average cost per call.
- Spend over time: a daily cost line across all agents and models. A rising line is rising spend.
- Spend by model: which models cost the most.
- Top spenders by agent: the agents driving the bill, each with a one-click "Set spend limit" that opens a pre-filled per-agent budget in monitor mode.
Spend appears once your agents report cost. The model-I/O guardrail path and the
AI Gateway report it automatically; with the SDK, pass model
and token counts (or costUsd) on the call.
Workspace budget
Set one monthly budget for the workspace. It carries an enforcement mode, so you choose how hard the cap bites:
| Mode | Behavior at the cap |
|---|---|
monitor | Track spend against the cap and surface it. No call is ever blocked. |
warn | Same as monitor, plus an alert when spend crosses the cap (80% is flagged on the way up). |
enforce | New governed calls are held for approval once month-to-date spend reaches the cap, until the next cycle or a higher cap. |
Enforcement never weakens risk governance
The budget's enforce mode can only add a hold on top of your risk verdicts. It
never relaxes them: a call your policies would deny is still denied, budget or
not. Spend enforcement tightens, it does not loosen.
Model-aware policies
Beyond the single workspace budget, policies can match on cost directly. A policy
context can gate on the model a call used or its callCostUsd, so "hold any call
to an expensive model" or "deny calls over $1 each" becomes one rule, evaluated
inline with the rest of your policies.
Catching spikes
When spend jumps relative to its recent baseline, a spend-spike recommendation
surfaces on the dashboard with a one-click path to a budget or a model-aware
policy, so a cost anomaly becomes an action rather than a surprise on the invoice.
Plans
| Capability | Plan |
|---|---|
The /spend cost dashboard and cost-anomaly alerts (spendGovernance) | Pro and above |
Workspace budget enforcement and model-aware cost policies (spendEnforcement) | Team and above |
Pro sees the spend; Team can act on it. (Free retains 7 days of history, so the dashboard is a Pro and above feature.)
Next steps
- Agent budgets: hard per-agent caps, enforced inline.
- AI Gateway: meter spend across every model call with one base URL.
- Policies: turn cost signals into verdicts.