ClovieEvery token and GPU-hour, priced and owned.
Every request is priced at the rates you actually pay and attributed to an agent, team and project. Budgets are checked before the call, so a runaway loop stops at the limit instead of on the invoice.
Costs
| Contact Centre$61,900 of $70,000 · hard stop | At risk |
| AML & Financial Crime$148,200 of $150,000 · monitor | Forecast to exceed |
| Trade Surveillance Analyst$41,000 of $40,000 · hard stop | Exceeded |
| Mortgage Lending$38,700 of $60,000 · notify | On track |
| Time | Agent | Model | Price book | Cost |
|---|---|---|---|---|
| 09:41 | Fraud Alert Triage | claude-sonnet-4.5 | Anthropic agreement −18% | $0.0184 |
| 09:41 | SQL Analytics Copilot | llama-3.3-70b · self-hosted | GPU time $3.85 / h | $0.0061 |
| 09:40 | Digital Banking Chat Assistant | gpt-4.1 | Azure agreement −22% | $0.0127 |
AI invoices arrive per account, not per owner.
No attribution
Provider bills don’t say which agent, team or product spent the money.
List prices lie
Negotiated discounts, commitments and self-hosted GPUs make list-price dashboards wrong.
Alerts after the fact
A monthly alert doesn’t stop a loop that spends a quarter’s budget overnight.
Clovie puts the budget check in the request path and the price book in the ledger, so finance and engineering read the same number.
Every request is priced before and after it runs, at the rates you actually pay.
See it at work
Register agent
- Name
- Card Dispute Triage
- Owner team
- Cards & Payments
- Project
- Customer Channels
- Environment
- Development
- Monthly budget
- $2,000
Costs
| Agent | Team | Tokens | Cost |
|---|---|---|---|
| Card Dispute Triagethis request, blocked | Cards & Payments | 0 | $0.00 |
| Fraud Alert Triage | Cards & Payments | 18,420 | $0.31 |
| KYC Document Review | KYC Operations | 42,905 | $0.74 |
What you get
Budgets with real stops
Notify, throttle to the lowest-cost model, or hard stop, for the organization, teams, projects and agents. Every budget starts in monitor mode.
Limits per request and run
Cap a single request, a single agent run or a short time window.
Fixed, planned or auto-adjusting
Budgets can stay fixed, follow a plan per period, or follow recent spend within floors and ceilings.
Negotiated prices
Fixed rates or discounts per model and region; reprice past usage when a contract changes.
Time-based and GPU pricing
Price provisioned throughput or self-hosted GPUs by time, shared fairly between concurrent calls.
Allocation
Cost categories, shared-cost splits and cost centers decide who pays.
Also in this layer
| Capability | What it does |
|---|---|
| Request tags | Tag requests with your own metadata for chargeback, budgets and policies. |
| Statements | Monthly chargeback and showback statements per team and cost center. |
| Anomalies and per-person usage | Review days when spend jumped, and see usage and spend per person and team. |
Controls mapped to the frameworks you answer to
- SOC 2: SOC 2 Type II trust services criteria
- NIST AI RMF: NIST AI Risk Management Framework
- ISO/IEC 42001: ISO/IEC 42001 AI management system
Runs wherever Clovie runs
- SaaS
- Customer VPC
- On-premises
- Air-gapped
Cost data and budgets live in your install in self-hosted modes. The model price catalog updates with each release in air-gapped installs.
Questions
Does Clovie bill us for model usage?
No. Providers bill you directly for requests made with your keys. Clovie prices, tracks and allocates those costs.
How is a budget checked before the cost is known?
The check runs before the call, so a hard stop refuses an over-budget request before it reaches the provider; the actual cost is recorded in the ledger after it.
What happens to failed calls and cached answers?
They cost $0, and gateway overhead is never charged.
Can a team get a temporary increase?
Yes. Team leads can request one, and every budget change is in the history with who, when and why.
Put every dollar of AI spend under an owner.
See cost controls and finops on your own agents, in the deployment you need.